EH_2024

EH_2024_04

Vol 33 ISSUE 4/24 • November 2024 THE EUROPEAN FORUM FOR THOSE IN THE BUSINESS OF MAKING HEALTHCARE WORK THE EUROPEAN FORUM FOR THOSE IN THE BUSINESS OF MAKING HEALTHCARE WORK AI imaging More accuracy in the segmentation of brain tumours Diagnosing cancer and ma- naging a patient’s respective treatment path requires a pre- cise segmentation of the af- fected anatomical structures. Defining the different sem- antic objects in an image such as disease patterns, lesions, biomarkers, organs, tissues etc. is at the core of this. Such a segmentation enables radi- ologists to distinguish the three subcategories of a tu- mour – the active core, the ne- crotic disintegrated area and oedemas – among others. A range of medical decisions are based on this classification: e.g., radiologists determine the volume of a tumour, moni- tor its development, design the concept of a personalised radiation therapy and sub- sequently administer it. In the realm of surgery, image seg- mentation is used for plan- ning and navigating oper- ations. To best support patients, clini- cians need this segmentation to be as accurate as possible. How- ever, segmenting medical images is a key challenge of medical image processing. The Research Department Artificial Intelli- gence in Medical Image and Sig- nal Processing (AIMedI) at the Lübeck site of the German Re- search Center for Artificial Intel- ligence (Deutsches Forschungs- zentrum Künstliche Intelligenz or DFKI) has been working on the development of a high-precision three-dimen- sional viewer software for medi- cal image data. Using AI and deep learning methods, the soft- ware analyses medical image data, bio signals along with other patient data and automatically interprets them. In this way, it can delineate pathologically al- tered image areas and detect dis- eases. für Prof Dr Heinz Handels, Director of the Institute of Medical In- formatics, University of Lübeck, and Head of AIMedI explains: “Our aim is to support clinicians with the classification of cancer tissue of malignant brain tu- mours such as glioblastoma. This also helps with the volumetric measurement of a brain tumour for which we need to be able to define which pixels form an ob- ject.” www.healthcare-in-europe.com Heinz Handels Prof Dr Heinz Handels Director of the Institute of Medical Informatics, University of Lübeck adversarial matically generated, using gener- ated net- works (GANs) or diffusion models.” However, this type of network poses its own chal- lenges like its substantial storage requirements. synthetic investigate Furthermore, data would circumnavigate data pro- tection requirements as man- dated per the General Data Pro- “In our tection Regulation. research, we to use only synthetic data for the training of neural networks. This will be of enormous value for re- search purposes as this data will be freely available”, concludes Handels. This is a relatively new research field, and more work is needed. ■ Grün: Peritumorales Edem; Gelb: Gadolinium-verstärkter Tumor k c e b ü L f o y t i s r e v i n U © Annotated medical data in short supply Methods to measure the volume of tumours already exist, but their accuracy is still wanting. To this end, the research group around Handels, AIMedI, devel- oped a procedure that allows to automatically segment and re- cognise objects such as tumours at an exact pixel-level in the brain. It uses deep learning methods tailored to image pro- cessing, which deliver a higher accuracy of the segmentation. These deep learning methods au- tonomously learn from training data. To facilitate this, the al- gorithm requires original images that must be manually marked by an expert to create a so-called annotated image. This approach has already been successfully used in many areas of public life, such as for the analysis of traffic, where non- the an- specialists perform notation of the images. Neverthe- less, this is different in the field of medicine where an expert is required to do the annotation. As it takes several hours to annotate a 3D dataset, only few exist: “I reckon that in my research com- munity there is globally only a li- mited number of such annotated datasets, which are all part of a global initiative. The data is held on central servers so that the im- ages can be accessed from every- where. This allows to compare generated images to those an- notated by experts”, says Han- dels. However, deep learning al- gorithms deliver more reliable results, the more data is avail- able. Using U-Nets for optimised results To overcome the constraints of limited manually annotated train- ing data, scientists in the AIMedI group have trained a U-Net, a convolutional neural network seg- developed image for training mentation. The U-Net architec- ture is suited to work with fewer (annotated) images whilst achieving a more precise segmentation. Furthermore, the researchers investigate the safety and possible explanations of the generated results. The evaluation of the methods and systems developed is still ongoing and done in close co- operation with medical co- operation partners based on practical applications. The future: Generating annotated synthetic image data At the heart of this endeavour is the aspiration to generate an- notated synthetic brain tumor images to free up experts’ time, which they anyhow only unwill- ingly bestow on annotating 3D datasets. Handels explains: “We aim to generate new data, so- called synthetic image data, where the labeled data is auto- Article: Cornelia Wels-Maug CONTENTS RADIOLOGY LABORATORY AI RESEARCH 2 5 7 11 www.healthcare-in-europe.com

2 RADIOLOGY PACS administrators as quality control monitors of algorithms The “watchdogs” of AI radiology tools m o c . e b o d a . k c o t s - a b p o p © i Artificial intelligence (AI) prod- ucts are starting to demonstrate that they can improve efficiency and quality in medical imaging. Radiology software vendors are rapidly developing AI tools, and government regulatory agencies are starting to clear them for clinical use. Radiology AI products are a whole new world. So is running them safely and efficiently in produc- tion. However, radiology AI quality control “watchdog” tools for monitoring and assessment have not been deployed as rapidly. New skills and resources are needed to use radiology AI tools safely and cost effectively, Raym Geis, MD, told attendees of the Samuel J. Dwyer III Memorial Lecture, the most prestigious event of the 2024 Society for Im- aging Informatics in Medicine (SIIM) Annual Meeting. Avoiding the “Challenger Shuttle moment” He warned that as radiology AI tools proliferate and begin to automatically communicate with each other, making independent decisions based upon their inter- actions, they will quickly evolve into highly complex models. These will require sophisticated, quantitative monitoring to prevent human injury, or as Geis calls this, “the Challenger Shuttle moment”. If an incident in which AI tools were to cause harm to many people, clinical medical imaging AI would come to an abrupt halt, pending lengthy investigation. Geis’ lecture, entitled “Reliability Engineering for Intelligent Medi- cal Imaging Systems”, suggested that radiology Picture Archiving and Communication Systems’ (PACS) administrators are best qualified to perform this role, but will need to learn challenging and sophisticated new skills. He be- lieves SIIM should take the lead in this new initiative, as it is the pro- fessional medical society that pioneered radiology PACS devel- opment and implementation. Geis, a radiologist and a radiology informatics specialist, Adjunct As- sociate Professor of Radiology at National Jewish Health in Denver, Colorado, and Clinical Assistant Professor of Radiology at the Uni- versity of Colorado School of Medicine, foresees a remarkable new future for SIIM: Leading the experts to make medical imaging AI into reliable, safe, efficient, and cost-effective clinical systems. ‘PACS administrators already do many things that AI will need: in- stallation, monitoring, manage- ment, repair and rollback. Radiol- ogy lags behind other industries in systems reliability engineering (SRE), but this works for us. We can adopt a lot of what other people in different industries have developed and are using,’ Geis said. ‘SIIM should take the lead to edu- cate current members on the prac- tice and methods of SRE as it applies to our specialized field, to promote and to define the re- search we need to understand, and to define best practices to op- erate these AI systems in medical imaging,’ emphasized Geis. ‘SIIM members’ strength is not to build these AI models, but to know how to manage these systems reliably so that there will be no medical catastrophe.’ The inevitable downward drift of AI tools The challenges are manyfold. ‘The initial one is that there is no way to predict which best AI tool will work over the long haul,’ he noted. ‘AI-modelled behaviour is determined by data. If the training data changes, the model changes. Accuracy will change. A PACS ad- ministrator doesn’t know which data features are the most critical,’ he explained. currently ‘What we do know is that any AI tool works better on the data it was trained on than the data it utilizes in a real-world clinical ap- plication. Unless the model is trained solely on your data, and even if you never change a tiny thing about how to generate the data, your results will be worse,’ said Geis. ‘The AI tool may work well on some of your exams, but not all. But you won’t know until you try it and then evaluate it.’ ‘Even if initially perfect, your AI tool’s results will get worse over time. Usually this is because the input data changes. Your depart- ment changes techniques and protocols. The patient population changes. The diseases change. A new scanner has been purchased that produces slightly different data. Who knows what else You need to be observant through monitoring. All of these changes are ultimately going to happen and your results are going to drift downward. AI tools fail silently, slowly, and subtly.’ Higher stakes Geis noted that if an AI model used by Google fails, advertisers may lose money. But nobody dies. Intelligent to grasp the emerging opportunity to become well paid Reliable Sys- tems Engineers. Compared to other professional organizations, SIIM is positioned develop a Certi- fied In- formatics Professional credential.’ He concluded, ‘This is a really ex- citing time. SIIM needs to grasp the opportunity and rejuvenate, to again become the leader and pion- eering medical society of AI in radiology.’ ■ Imaging Report: Cynthia E. Keen But in radiology, rigorous moni- toring is essential. In his lecture, Geis described the steps to evalu- ate, adopt, monitor, retrain, and decommission an AI model. He emphasized the importance of understanding and being able to access the data a vendor used to train the model, and the import- ance of logging versions of both the vendor software model and data set(s) tested. He explained the concept of “banded assess- ment monitoring”, where specific groups of features are assessed for QC and performance accuracy. Some questions for evaluation in- clude: What is the accuracy level stated by the vendor and how does it compare Is AI accuracy consistent among lo- cations, different models/vendors of an imaging modality What is the impact of different radio- therapists performing the exam, or radiologists it How does an AI tool interpret atypical exams interpreting imaging Setting the standards Geis exhorted SIIM as an organi- sation to develop model card tem- plates for medical imaging. In other industries, model cards con- tain model number, use case, architecture, information about training data and how it is evalu- ated, ethical considerations when the AI tool should not be used, and contact information to enable direct contact with the creators. He urged SIIM to set the standard for vendor AI software tool model cards, and to be the preeminent decider of information needed for monitoring to make AI tools safe and reliable. ‘Monitoring is expensive. It is es- sential. PACS administrators need Raym Geis Raym Geis, M.D., a radiologist and imaging informaticist, is Adjunct As- sociate Professor at National Jewish Health in Denver, CO, and a Clinical Assistant Professor at the University of Colorado School of Medicine in Aurora. His interests include systems engineering for medical imaging AI, as well as radiology data, standards, and ethics of new data science ap- proaches for medical imaging. Geis is a past chair of the Society for Im- aging Information in Medicine and former Vice Chair of the American College of Radiology Informatics Committee. EUROPEAN HOSPITAL Vol 33 Issue 4/24

RADIOLOGY 3 GI Genius enters NAIAD large-scale clinical trial –Advertorial– AI making earlier detection of colorectal cancer possible The intelligent endoscopy module detects and marks precancerous lesions in real-time c i n o r t d e M © Early detection is crucial for ef- fective treatment of colorectal cancer, but the initial stages are easily missed during endoscopy screenings. With GI Genius, Med- tronic presents an AI-powered tool to help detect precancerous lesions in real-time. In an up- coming large-scale trial, the in- telligent endoscopy module will demonstrate its benefit in the clinical context. Colorectal cancer is among the most common in developed coun- tries and is responsible for more than 150,000 deaths per year in Eu- rope alone.1 On the other hand, 90% of patients can recover when the cancer is detected at an early stage.2 While small benign polyps can start to grow in the colon, they can become adenomas and from that, progress into cancer. established This process can take years, often without symptoms, so early detec- tion is essential. To prevent unde- tected progression, many countries have endoscopy screening programmes. However, since the initial polyps may be very small, and the procedure covers a considerable distance of the colon, abnormalities can be easily missed, even by experienced endoscopists. This is further exacerbated by an increasing workload in the special- ty. to support A second set of eyes – but not a re- placement for human skill To mitigate this problem, GI Gen- ius system acts like a “second set of eyes” gastro- enterologists during colorectal cancer screening. The system re- views the colonoscopy images, in real time3 and throughout the pro- cedure, highlights suspected re- gions that the clinician may use to confirm whether the area contains a lesion. This overlay marker may be accompanied by an additional AI assessment of the area, cat- egorizing it as either adenoma, non-adenoma or no prediction. The final decision remains with the expert endoscopist, but the tech- nology is designed to increase the level of confidence4 in the optical diagnosis, and, unlike the operator, whose vigilance may deteriorate after several hours of looking at similar images, the AI remains laser-focused on its task, suppor- ting the detection of colorectal polyps of various shapes, size, lo- cation and histology.5,6 The increased diagnostic con- fidence can also bring financial benefits. Current practice is to re- move almost every suspicious find- ing from the colon and send it to histopathology for further analysis, which can be time-consuming and a significant cost burden. The inter- national guidelines suggest that a highly accurate optical diagnosis should permit the implementation of cost saving strategies such as re- sect-and-discard and a diagnose and-leave. AI-assisted polyp char- acterization with high accuracy could expand the uptake of optical diagnosis, increasing the cost–ef- fectiveness of colonoscopy.7 Large-scale clinical study puts AI to the test GI Genius is the most studied AI system in endoscopy which so far proved its credentials in a series of studies: Research found an in- crease of adenoma detection rate of 14.4% and approx. 50% reduc- tion in adenoma miss rate when the system was used.8,9 Especially the detection of smaller lesions (≤ 5 mm) significantly when the AI assisted the endoscop- ists.8 The system also contributes to maintaining a consistency of care by increasing diagnostic confi- dence of less experienced examin- ers, another study found.10 improved ance and its impact on endo- scopists in a “real-world” clinical setting has kicked off in the UK. The National study of Artificial In- telligence in Adenoma Detection for colonoscopy (NAIAD) is led by King’s College London. The trial will be carried out over two years and eventually involve the deploy- ment of 60 AI systems across 20 hospitals in the UK National Health System (NHS) in England. The study is the result of set-up funding from NHS England, which ran a competition to identify and fast- track AI technology into clinical services.12 “Our NAIAD study is set to explore the use of AI in a ‘real world’ setting, and how it might in- fluence endoscopists in their day- to-day practice,” says Dr Bu Hayee, Consultant Gastroenterologist and Principal Investigator of the study at King’s College Hospital. The use of AI in endoscopies is set to expand as GI genius has the possibility to host multiple real- time applications which will sup- port development and access to these tools. AI in endoscopy has a bright future and has the potential to continue to improve patient out- comes. ■ Two new major pieces of research are underway to further evaluate GI Genius COLO-DETECT is a randomised control trial that evaluated whether colonoscopy assisted with the GI Genius improves the detection of abnormalities such as polyps and cancer compared to standard col- onoscopy. participants, across 11 sites in the UK, have been randomised to receive either 2,000 GI Genius assisted colonoscopy or standard colonoscopy. The study has found that GI Genius sig- nificantly increased the adenoma detection rate and increased sessile serrate polyp detection, overall, it recommends GI Genius to be in used in all colonoscopies.11 In addition, a large-scale study which was made possible by the NHS award, is designed to assess the system’s diagnostic perform- GI Genius™ intelligent endoscopy system Indications for use The GI Genius™ module is a computer-assisted reading tool designed to aid endoscopists in detecting colonic mucosal lesions (such as polyps and adenomas) in real time during standard white-light endoscopy examinations of patients undergoing screening and surveillance endoscopic mucosal evaluations. GI Genius™ computer-assisted detection device is limited for use with standard white-light endoscopy imaging only. This device is not indented to replace clinical decision-making. Risks If the device is used outside of Instructions For Use, it could potentially lead to inappropriate diagnostic information being displayed to the user (e.g. to analyze images from an unintended patient population, on images acquired with incompatible imaging hard- ware or incompatible image acquisition parameters). Incorrect detection or characterization of lesion(s) may lead to false positive or false negative which may result in incorrect patients’ management with possible adverse effects: Unnecessary treatment, unnecessary additional medical imaging and/or unnecessary additional diagnostic workup such as biopsy, complications, including incorrect diagnosis and delay in disease management. Device failure could lead to the absence of results, delay of results or incorrect results, which could likewise lead to inaccurate pa- tient assessment. In the event of unrecoverable failure, please switch off and revert to Non AI enhanced colonoscopy. Important: Always refer to the Instructions For Use (IFU) packaged with the product for complete instructions, indications, contrain- dications, warnings and precautions. ©2024 Medtronic. Medtronic, Medtronic logo and Engineering the extraordinary are trademarks of Medtronic. All other brands are trademarks of a Medtronic company. ™*Third-party brands are trademarks of their respective owners. emea-dg-2400043-gi-genius-ar- ticle-en-gb-emea-12571174 1 Colorectal Cancer burden in EU-27. ECIS Accessed 15th March 2024. Available at: https://ecis.jrc.ec.europa.eu/pdf/Colo rectal_cancer_factsheet-Mar_2021.pdf 2 Understanding Statistics Used to Guide Prognosis and Evaluate Treatment. ASCO. Accessed on 15th March 2024. Available at: https://www.cancer.net/navigating-cancer- care/cancer-basics/understanding-statistics- used-guide-prognosis-and-evaluate-treat- ment . 3 IFU v2.0.0 GI Genius FINAL 4 Hassan C, Balsamo G, Lorenzetti R, Zullo A, Antonelli G: Artificial Intelligence Allows Leaving-In-Situ Colorectal Polyps; Clinical Gastroenterology and Hepatology 2022; https://doi.org/10.1016/j.cgh.2022.04.045 5 Wu J et al.: Comparison of efficacy of colon- oscopy between the morning and afternoon: A systematic review and meta-analysis; Di- gestive and Liver Disease 2018; https://doi. org/10.1016/j.dld.2018.03.035 Medical Device for Endoscopy; Bioengineer- ing 2023; https://doi.org/10.3390/bioeng ineering10040404 7 Messmann H et al.: Expected value of ar- tificial intelligence in gastrointestinal endo- scopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement; Endo- scopy https://doi. org/10.1055/a-1950–5694 2022; org/10.1053/j.gastro.2020.04.062 9 Wallace MB et al.: Impact of Artificial Intel- ligence on Miss Rate of Colorectal Neoplasia; Gastroenterology 2022;163:295–304; https://doi.org/10.1053/j.gastro.2022.03.007 10 Repici, A et al.: Artificial Intelligence and colonoscopy experience: lessons from two randomized trials. Gut. 2021; https://doi. org/10.1136/gutjnl-2021–324471 6 Cherubini A, Dinh, N.N: A Review of the Technology, Training, and Assessment Methods for the First Real-Time AI-Enhanced 8 Repici A et al.: Efficacy of real-time com- puter-aided detection of colorectal neoplasia in a randomized trial; Gastroenterology. 2020. https://doi. 159:512–520.e7; 11 Seager A et al.: Polyp detection with colon- oscopy assisted by the GI Genius artificial in- telligence endoscopy module compared with www.healthcare-in-europe.com standard colonoscopy in routine colonoscopy practice (COLO-DETECT): a multicentre, open-label, parallel-arm, pragmatic rando- mised Lancet Gas- troenterology & Hepatology 2024; https://doi. org/10.1016/s2468–1253(24)00161–4 controlled trial; 12 King’s trials AI to aid early detection of bowel cancer; Kings College Hospitals NHS Trust website/news; https://www.kch.nhs.uk/ news/kings-trials-ai-to-aid-early-detection-of- bowel-cancer/ Last accessed February 2024

4 RADIOLOGY AI, modern mammography, and more EUSOBI 2024: Breast imaging breakthroughs Strategic Minimally invasive surgical inter- ventions, innovative imaging and the use of AI: At the EUSOBI con- gress in Lisbon, experts pres- ented and discussed the latest advances in breast imaging. We spoke with Tanja Brycker, Vice President, Devel- opment, Breast & Skeletal Health and Gynecological Surgical Sol- utions at Hologic, ahead of the event about new in women’s health, the company’s investment in innovation and education, and what the future of mammography like with the rise in AI. trends looks EH: What are the most significant trends in women’s health that you’ve observed recently, particularly in di- agnostic imaging, and how will the- se be highlighted at the congress? Tanja Brycker: ‘Minimally invasive surgical interventions for breast cancer, artificial intelligence (AI) contrast-enhanced mam- and mography (CEM) and biopsy (CEBx) are amongst the most sig- nificant current trends in women’s health. The increasing focus on mi- nimally inter- ventions for breast cancer is driven by the potential to improve patient outcomes and well-being, while also benefiting healthcare facilities through reduced demand on re- sources and a potential reduction in costs. invasive surgical ‘Diagnostic imaging is also no dif- ferent from other healthcare fields, with AI continuing to be a hot topic, despite having been in use for over a decade. While it is used to manage and clarify diagnostic images, the impact of AI also ex- tends to driving workload efficien- cies and the potential to support more personalized medicine. ‘Research has already shown that the next generation of deep-learn- ing technology, such as Hologic’s Genius AI Detection solution, can help significantly enhance per- formance, surpassing traditional machine learning Computer-Aided Detection (CAD) algorithms in spe- cificity and overall effectiveness.1 ‘Finally, to mention the growing focus on CEM and CEBx. The syn- ergy between CEM and CEBx has the potential to accelerate cancer diagnosis and workflow efficien- cies. CEM pairs 2D and tomo- synthesis images, all under one compression, providing anatomical and functional imaging in one exam.2 The contrast agent ac- cumulates where lesions are form- ing and growing, enabling radi- ologists the Hologic Selenia Dimensions and 3Dimensions systems. When a breast lesion is identified, CEBx en- ables radiologists to target and ac- identify it on to m o c . e b o d a . k c o t s – a l l e n u m a s © ports adoption of, and confidence in, the technology. Chief of Breast Imaging at Careggi University Hos- pital, Dr. Jacopo Nori, has been a great advocate for CEM education, helping EUSOBI attendees learn about CEM image interpretation, its use in pre-surgical planning, and other medical education work- shops. We’re proud to work with Dr. Nori on these sessions to build on last year’s activity at our EUSO- BI symposium. ‘Additionally, we continue to spon- sor the EUSOBI Breast Imaging Grant for ten Young EUSOBI Club members to attend the congress as part of our dedication to invest in the future of diagnostic imaging. We believe that future radiologists should have the opportunity to network with their peers and build relationships with industry leaders, while having access to the latest educational opportunities.’ Can you share any partnerships or collaborations your company has engaged in to foster innovation and education in the field, and were any of these highlighted at the event? is imaging. The ‘Through our collaboration with BARCO, we are excited to an- nounce the arrival of a new moni- tor for PACS and breast imaging this year. The Coronis Uniti moni- tor the biggest and most equipped display system yet avail- able for general radiology and breast full-size, 12-million-pixel screen presents images with crystal-clear precision on a large, flexible, bezel-free image area, providing an uninter- rupted view of large images, nu- merous smaller studies, or various combinations and overlays in any layout. EUSOBI congress attendees were able to see it on BARCO’s booth.’ quire tissue samples with the same contrast agent. versatility to sample larger vol- ume tissue specimens. ‘These three trends are reflective of the growing awareness of how technology can help address radi- ologist workload, drive workflow efficiencies and improve patient comfort and satisfaction. By de- livering technologies that support diagnostic accuracy, we can im- prove the experience of everyone involved in the breast diagnostic imaging process.’ What did Hologic showcase at the EUSOBI congress? ‘Hologic highlighted a series of in- novations designed to help radi- ologists improve workflow effi- ciency and diagnostic accuracy – from AI solutions designed specifi- cally for breast cancer diagnosis, to expanded breast biopsy solutions. Attendees could visit Hologic’s booth at the Lisbon Conference Centre to explore: • Contrast-enhanced Mam- mography Solutions that enable clinicians to identify and target lesions that are not visible in 2D or 3D mammography images. • Hologic‘s integrated Breast Health AI Software Suite that im- proves early detection and diag- nosis of breast cancer, helps to accelerate workflow, and enables personalized patient care with automated breast density assess- ment. ‘Additionally, we provided a com- plete medical education pro- gram featuring wide-ranging read- ing and interventional workshops that provide participants with hands-on experiences. After last year’s successful CEM education workshops, Dr. Jacopo Nori, chief of breast imaging at Careggi Uni- versity Hospital, returned to lead another series on CEM with his team. He was also joined by Dr Julia Camps-Herrero, Head of the Radiology Department at Univer- sity Hospital de la Ribera in Alzira, Valencia, Spain, to discuss new re- search about using CEM in manag- ing patients with a personal history of breast cancer. imaging ‘We also welcomed Dr. Sarah Frie- dewald, associate professor of Radiology and chief of breast im- aging at Northwestern Memorial Hospital, for an interactive hands- on reading workshop with our 3DQuorum technology. Additionally, Dr. Adnan Duhovic, radiologist at Goldenes Kreuz Pri- vate Hospital in Vienna, Austria, shared his experience and learn- ings tomo- synthesis (DBT) and the best use of the technology in a nationwide screening programme.’ for digital breast Why does Hologic specifically in- vest in education within the diag- nostic imaging sector? • The Brevera Breast Biopsy Sys- tem, the world’s first vacuum- assisted breast biopsy solution that integrates tissue acquisition, real-time imaging and verifi- cation, and post-biopsy hand- ling. With the newly launched 7-gauge needle, the system now empowers pro- fessionals with enhanced clinical radiology for radiologists ‘At Hologic, we have a long- standing commitment to invest- ment in clinical education. It is es- sential to understand how to correctly use a system for optimized performance and accuracy. CEM is a great example of this, and of the sig- nificant investment we have made in an education strategy that sup- 1 Performance of a traditional machine learning computer-assisted detection (CADe) algorithm versus a deep learning artificial in- telligence (AI) algorithm on digital breast to- mosynthesis (DBT) studies. Authors: Mani- sha Bahl (MGH), Constance Lehman (MGH). Accepted as electronic poster. 2 Burhenne LJW., Wood SA., D‘Orsi CJ., et al. Potential Contribution of Computer-aided Detection to the Sensitivity of Screening Mammography, 2000; 215:554–562 Radiology 3 Based on analyses that do not control type I error and therefore cannot be generalized to specific comparisons outside this par- ticular study. In this study: The average ob- served AUC was 0.825 (95% CI: 0.783, 0.867) with CAD and 0.794 (95% CI: 0.748, 0.840) without CAD. The difference in observed AUC was +0.031 (95% CI: 0.012, 0.051). The average observed reader sensitivity for cancer cases was 75.9% with CAD and 66.8% without CAD. The difference in observed sensitivity was +9.0% (99% CI: 6.0%, 12.1%). The average observed recall rate for non- cancer cases was 25.8% with CAD and 23.4% without CAD. The observed difference in negative recall rate was +2.4% (99% CI: 0.7%, 4.2%). The average observed case read-time was 52.0s with CAD and 46.3s without CAD. How is AI transforming the field of mammography, and what specific advancements has Hologic made in this area? Was this also showcased at the EUSOBI congress? is ‘AI revolutionizing mam- mography by enhancing early de- tection and diagnostic accuracy, as- sisting radiologists with identifying subtle abnormalities with greater confidence. As we look at AI today and into the future, it can help em- power clinicians to deliver more personalized patient care. ‘As a global leader in breast health, we see AI playing a major future role in breast imaging. Because of this, we continue to make sig- nificant investments in research and innovation related to AI. Our suite of AI solutions is reflective of this, and can help radiologists with optimizing their work, including triaging patients, reading images, and planning for surgery. At our booth at EUSOBI, we did showcase this suite of integrated breast health AI solutions for attendees to explore. ‘Most notably, we introduced atten- dees to our new Genius AI Detec- tion solution – which can help radiologists categorize and priorit- ize cases by complexity, to opti- mize workflow and expedite pa- tient care. This new deep-learning AI software is designed to help radiologists detect subtle potential cancers in breast tomosynthesis images. With higher specificity than Hologic’s previous AI sol- ution, the deep-learning software can help detect hard to identify lesions for further examination by the radiologist. Research shows a difference of +9% in observed reader sensitivity for cancer cases using Genius AI Detection tech- nology.3 overlapping ‘Also at our booth was our 3DQuo- rum imaging technology, which as- sists radiologists by reducing the number of DBT images they need review by creating high- to resolution, 6mm „SmartSlices“ to expedite reading time. When a radiologist reads SmartSlices instead of 1mm slices, the number of DBT images to re- view is reduced by two-thirds3–5, leading inter- pretation time savings of one hour per day.4 to an average ‘AI will continue to revolutionize mammography for years to come and will help improve the radi- ologist experience. Our suite of in- tegrated AI solutions can help fa- cilitate the workflow efficiency and diagnostic accuracy that they are looking for.’ ■ Interview: Wolfgang Behrends The observed difference in read-time was 5.7s (95% CI: 4.9s to 6.4s). 4 Data on File: Clinical Study Report CSR-00116 Rev. 004 EUROPEAN HOSPITAL Vol 33 Issue 4/24

LABORATORY Molecular meets digital 5 Single-cell multiplex imaging: a powerful tool for digital pathology tween cells. ‘It is a more ad- vanced way of doing image analysis. The difference in tissue research is that you can map those phenotypic clusters and cells on top of the tissue as we have coordinates for every single segmented and classified cell and that gives a lot of data as well.’ The FIMM team has recently launched a pan-cancer multiplex tumour microenvironment pro- ject aiming to profile over 40,000 tissue samples. By analysing samples, survival data and patho- logical data, they are using five different antibody panels to cover 33 different tumour micro- environment markers to discover both ‘unique and common cellu- lar profiles in these cancers.’ they state. The importance of spatial context He gave an example of identified multimarker-defined fibroblast subsets in non-small cell lung cancer with different associations of patient survival, mutations and immune features, highlighting CAF (cancer-associated fibrob- last) subsets with some showing better prognosis and survivability than others. ‘The study high- lighted that there are different types of fibroblast subsets and also suggests that different CAF subsets can have tumour per- mitting and tumour inhibiting functions in cancer,’ said Pel- linen. The team also examined the prostate cancer tumour micro- environment, focusing on the be- haviour and density of CD3 im- mune cells in benign and tumour areas and discovered highly in- teresting spatial interactions be- tween CD3+ immune cells and epithelial cells with prognostic significance. ‘This example,’ he said, ‘highlights the importance of spatial context in cancer and really tells us that large scale im- munofluorescence profiling of cancer can reveal spatial patterns with highly interesting clinical associations that would be missed with technologies not preserving the tissue integrity.’ Transformative potential of liquid biopsy In her presentation, molecular pathologist Dr Heather Dawson from the Institute of Tissue Medi- cine and Pathology at the Univer- sity of Bern, Switzerland, said that while novel technologies have the potential to transform, transition from basic research to clinical practice can take a long time. Focusing on the example of liquid biopsy to illustrate a tech- nology with transformative po- tential for healthcare, Dawson said the technique can be used to measure markers such as circu- lating tumour cells, exosomes and circulating tumour DNA (ctDNA). Illustrating the point through the patient journey, she highlighted how liquid biopsy can be used for early diagnosis of cancer, minimal residual disease (MRD) after an intervention, therapy monitoring, and detection of mu- tations and resistance mech- anisms in advanced disease. She said: ‘The introduction of liquid biopsy, especially for the ap- plication of early cancer diag- nosis and MRD, opens the door for molecular testing in patients with localised disease.’ ■ Report: Mark Nicholls Teijo Pellinen Dr Teijo Pellinen is a Senior Univer- sity Researcher at the Institute for Molecular Medicine Finland (FIMM) at the University of Helsinki, where he leads team focusing on and devel- oping advanced imaging technol- ogies, especially on cancer specimens. Heather Dawson Dr Heather Dawson is Medical Di- rector of Molecular Pathology at the Institute of Tissue Medicine and Pa- thology at the University of Bern in Switzerland. Her research focus is the investigation and use of prognostic biomarkers in the diagnosis of colo- rectal carcinoma. information Multiplex imaging can play a critical role in unravelling the tumour microenvironment. The potential and benefits of the emerging approach – a way to extract from human tissue samples by vis- ualising many more biom- arkers than traditional micro- scopy – was highlighted in presentations during the 36th European Congress of Pathol- ogy in Florence, Italy. Speakers also discussed novel tech- nologies to transform clinical care and deep learning for pa- thology biomarkers for preci- sion oncology. indicated In his presentation, Dr Teijo Pel- linen that under- standing the diverse cell states and molecular composition with- in the tumour microenvironment is crucial for improving risk strat- ification and predicting therapy responses. The Senior University Researcher at the Institute for Molecular Medicine Finland (FIMM) at the University of Hel- sinki, Finland, said multiplex tis- sue imaging is ‘a great tool’ for discovering ‘clinically meaningful associations’ and can answer dif- ferent scientific questions. and simple Explaining the digital image pro- cessing and analysis steps in multiplex tissue profiling, he dis- cussed different spatial tissue im- aging approaches, from IHC- based systems, immunofluorescence, mass spec- trometry and tyramide signal am- plification systems. ‘Some are more high- throughput, meaning that they can cover large numbers of samples, while others are more suited to deep biological profil- ing and can cover up to 100 pro- tein targets but are typically li- mited to fewer sample numbers with expensive infrastructure costs,’ said Pellinen. ‘Basically, the choice of multiplexing sys- tem depends on the biological and scientific questions.’ A more advanced way of doing image analysis Pellinen highlighted how his FIMM group has developed an in-house approach that combines tryamide signal amplification and cyclic immunofluorescence in an affordable high-throughput sys- tem using basic histo-laboratory commercially fluor- escence scanners. With cyclic staining and digitisation, ma- chine learning (ML) is used to an- notate the artefacts, followed by cell segmentation and classifi- cation, though human input is still required, and multiple chal- lenges remain to improve the automation. available He said multiplexing offers more accurate phenotyping and the ability to measure distances be- www.healthcare-in-europe.com Trusted to Capture Every Image: It Has to be Aquasonic® 100 For generations, Aquasonic 100 has been the preferred transmission gel for equipment manufacturers and medical professionals alike. From prenatal screening to advanced diagnostics, it stands alone as the choice for clear and consistent scanning. Trust Aquasonic 100 to ensure you capture the images that matter most. Request a sample: ©2024 Parker Laboratories, Inc. The sound choice in patient care is a trademark of Parker Laboratories, Inc. AD-01-0032 REV 0

6 LABORATORY Keynote on integrated diagnostics Predicting – and shaping – the future of modern pathology is using tools to identify every po- tential member of the tumour microenvironment. ‘An important component of what we do in the future is associated with clinical trials, but until now clinical trials have been the lost opportunity to develop biomarkers meaningfully,’ he suggested. ‘The trend we see is that we probably apply AI too late in the process. It may make sense to bring this forward and start gen- erating AI tests up front.’ A critical step in delivering inte- grative diagnostics into hospital settings and generate new al- gorithms to help patients involves accessing information in real life in the hospitals, the expert pointed out. ‘The reason we are not able to extract information for patients im- mediately and start applying these tools is because that information is in siloes – lab information manage- ment systems (LIMS), electronic health records (EHR), other rec- ords, images, are not talking to each other. That is the main bottle- neck.’ ‘Custodians of multiplex diagnostic opinion’ A survey of clinicians suggested genomics, histopathology, digital pathology tools, AI results, radiol- ogy and radiomics, EHR and LIMS should be integrated. However, three main questions must be answered to overcome this chal- lenge, Salto-Tellez said: • are we ready in our hospital to manage that information in that way • how can • do we have computation tech- nology to create that integration be brought to a test that is an amal- gamation of different systems accreditation Complex diseases could require complex biomarkers for accurate diagnosis in the years ahead, ac- cording to a leading pathologist. In a keynote address to the 36th European Congress of Pathology in Florence, Italy, Professor Ma- nuel Salto-Tellez highlighted this as one of the major future chal- lenges for the discipline. How- ever, he also believes pathol- ogists will sit at the core of modern medicine. As he opened his address, the Pro- fessor of Integrative Pathology at the Institute for Cancer Research in London (ICR) and Professor of Molecular Pathology at Queen’s University Belfast (QUB) reflected on the last two decades looking in pathology and how at the turn of the millennium, he suspected few would have been able to predict the advances, and the speed at which they have been made, in the intervening years. In his presentation, Salto-Tellez also referred to the challenge in 1999 from Dr Harold Vamus, Nobel Prize winner, pioneer of oncology and Director of the US National In- stitute of Health, for the scientific community to change the basis of tumour classification ‘from mor- phological to molecular char- acteristics.’ Despite the pessimism it created then about the future of the discipline, Vamus said: ‘Could we have predicted that 25 years later, there were more samples than ever coming into pathology departments with an increase of 5% per year and that many of the predictive markers we used in personalised medicine are actually driven by good old immunohistochemistry ’ Disease vs biomarker complexity: levelling the playing field ‘So, how do we start predicting the future ’ asked Salto-Tellez. While there are new drugs, clinical trials and technologies, he focused on new biomarkers as pivotal in trans- lating discovery into applicability and prediction of treatment re- sponse. Focusing on immuno-oncology (IO) and advances in pathology, particularly with the advent of ar- tificial intelligence (AI), he dis- cussed the complex biology of the tumour microenvironment, the complex biomarkers and what pathologists can offer today in IO diagnostics with PD-L1, micro- satellite instability, tumour muta- tional burden, and – in the bound- ary between clinical trials and diagnostics – the analysis of T-cell response and tumour infiltrating lymphocytes (TILS). ‘We are deal- ing with complex disease with very simple biomarkers,’ he said. ‘But it could be that complex disease need complex biomarkers. To me, this is going to be that challenge that we are going to see over the next few years.’ Advancing that can mean extrac- ting the complexity that exists in images; harnessing new tech- nologies; or better use of in- formation available in hospitals and ‘begin to integrate that in new algorithms that are hopefully going to better help our patients.’ The future is digital He pointed to studies in digital pa- thology and AI that ‘tell us that a pathology service that is digitised is better. But this is only the begin- ning because we know that the in- silico image is going to allow us to extract significant information for the future.’ Such tools, however, need to be in- tuitive to use, meaningful for the pathologist and scientist, quanti- tative, and correlate with clinical outcomes. The AI architecture, with numerous tools now available to pathologists, can be used to identify histological features better, predict molecular biomarkers and quantitate single and multiple biomarkers. New methods against PD-L1 ‘madness’ Prof Salto-Tellez also reflected on the current complex pathology pathway in the realms of molecular diagnostics. ‘It is fragmented and as pathologists, we probably need to start thinking about how we rationalise this.’ He expanded on this by referring to the ‘madness’ with PD-L1, where pathologists are dealing with multiple scoring sys- tems associated with individual drugs, antibodies and auto-stainers which have to be scored and de- fined differently, depending on the cancer type. His QUB team is working to create prototypes to match the H&E and PD-L1 staining to mark the area of tumour and then render a potential result. When translated into an AI in a system together with other po- tential prognosticators, he said that can begin to tell the prognosis of individual patients. ‘Through the tools we have today, we are begin- ning to generate new complexity,’ he said. Medical information talks – if we let it Further ICR work, he said, has examined tens of clinical trials and NEWS ∙ REPORTS ∙ INTERVIEWS ∙ VIDEOS ∙ SOCIAL MEDIA S a y i n f rm e d ‘This is important because to me, the main custodian of that multi- plex diagnostic opinion should be the pathologist,’ said Salto-Tellez. ‘We should be the ones leading this opinion. My prediction is that we are looking at a fourth revolution: how to integrate our resources and how we bring that information into our patients is going to dictate the future.’ ■ Report: Mark Nicholls Manuel Salto-Tellez Professor Manuel Salto-Tellez is Professor of Integrative Pathology at the Institute for Cancer Research in London (ICR), and the lead of the Royal Marsden Hospital/ICR Inte- grated Pathology Unit. He is also the Chair of Molecular Pathology at Queen’s University Belfast (QUB), a clinical consultant pathologist and the Lead of QUB’s Precision Medicine Centre of Excellence, and author or co-author of more than 320 inter- nationally peer-reviewed articles in translational science, molecular pa- thology and diagnostics. EUROPEAN HOSPITAL Vol 33 Issue 4/24

AI Generative AI 7 Large language models: enabler or eroder of cardiovascular care age’, he pointed to an alternative view that overuse of LLMs ‘will lead to the destruction of current scientific process.’ Prof Tomasz Guzik is a Regius Chair of Physiology and Cardiovascular Pa- thobiology of the University of Glas- gow and a Professor of Medicine at Jagiellonian University, Poland. His research focuses on vascular biology, hypertension, and cardiovascular im- munology. and some suggests Evidence re- searchers are using LLMs to pre- pare scientific literature and he felt this can improve grammar, consistency formatting, streamline literature searches and reference management, and may also increase objectivity and re- duce bias. Guzik, who is Chair of Cardiovascular Medicine at the University of Edinburgh, added that it also facilitates continuous learning, data analysis and inter- pretation. But there are risks too, he pointed out. These include loss of originality and authenticity in content, misinterpretation of data visualisations, and overuse may remove need for collabor- ation. It may, he continued, also lead to the introduction and per- petuation of biases, and an over- reliance on AI may lead to an ero- sion of critical thinking. ■ Article: Mark Nicholls m o c . e b o d a . k c o t s - c i n o T h c e T © cation, he believes there is a huge opportunity for them to improve education. Impact on scientific publishing Tomasz Guzik, Editor-in-Chief of the journal Cardiovascular Re- search, looked at the challenges and opportunities of LLMs in scientific publishing. Noting that scientific research should be im- pactful, he explained that ‘in order to be impactful, research needs to be visible. LLM and AI can help authors to achieve this.’ While highlighting a view that AI could foster integrity and help re- the digital searchers ‘navigate F i n d a l p r o d u c s a l s n l i n e For more information, visit healthcare-in-europe.com Large language models (LLMs) have potential in healthcare set- tings to help support both pa- tients and clinicians. Cardiolog- ist Dr Robert van der Boon believes they could have several applications, including patient communication and education, clinical decision support and administrative tasks. Delegates to ESC 2024 in London heard roles explored for LLMs in areas of clinical decision- making, patient care, research, and scientific writing. But ex- perts in the session also raised concerns over trustworthiness, bias, privacy and security. vide incorrect or incomplete in- formation – and thus have a detri- mental effect. ‘One way we tackle this problem is to implement human oversight instead of let- ting LLMs answer patients di- rectly,’ van der Boon said. A recent study showed that LLM- powered chatbots can improve physician efficiency and inter- physician agreement. He also pointed to research that sug- gested LLMs may even be more empathetic clinicians, though the results of the studies should be interpreted with cau- tion. than Van der Boon, from the Cardiov- ascular Institute at the Erasmus Medical Center in Rotterdam, gave a specific presentation on how to improve patient care with LLMs. He said: ‘As healthcare has become very data and text heavy, LLMs have the potential with natural language processing capa- Robert van der Boon Dr Robert van der Boon is a car- diologist at the Cardiovascular Insti- tute at the Erasmus Medical Center in Rotterdam, the Netherlands. His research interests are in heart failure, telemonitoring and digital health. bility to improve efficiency but also provide better outcomes for our patients.’ For patients, one of the most ap- pealing options interaction with LLM-powered chatbots, but there is risk that these models may start to “hallucinate” – pro- is www.healthcare-in-europe.com A future goal would be to devel- op LLM-based chatbots enabling patients to talk to their doctor in the form of an Avatar 24/7. ‘But we are not there yet,’ the expert added. Reducing administration The technology also shows prom- ise in regards of diagnostics skills, van der Boon said: ‘LLMs do fairly well on exams – but we know from medical school that answering questions is not the same as being a doctor.’ Instead, he explained, a key area of bene- fit lies in reducing administrative tasks for doctors, particularly in an outpatient clinic where one hour of clinic creates up to two hours of administration, leading to stress and burnout in phys- icians. to barriers However, im- plementation such as accuracy of models, trustworthiness, bias, se- curity and privacy remain. The ex- pert said: ‘We need to prove the safety and effectiveness of these models in pragmatic, prospective randomised controlled trials be- fore widespread implementation. LLMs have the potential to revol- utionise patient care and will likely expand in the near future, Stéphane Fournier Dr Stéphane Fournier is an Attend- ing Physician at the Interventional Cardiology Unit, in the Department of Cardiology at CHUV in Lausanne, Switzerland. His research is primarily focused on coronary physiology but also includes the implementation of large language models (LLMs) in clinical practice. but challenges remain and inte- gration into clinical practice requires careful consideration.’ Potential for education During the session, Dr Stephane Fournier, Head of the Cath Lab at University Hospital Centre Vau- dois (UHCV) in Lausanne, Swit- zerland, discussed the impact of the technology on cardiovascular education. He elaborated on how LLMs offer better and equitable access to information, enable ‘shy’ students to ask questions without feeling uncomfortable in front of peers, are time efficient, deliver updated lan- guage assistance, and facilitate exam preparation and can even create clinical skill scenarios. information and However, he said a risk was that students would lose the ability to communicate, and posed the question: ‘Can someone who knows everything but can no longer communicate still be a good doctor We have to keep in mind that the communication with our patients and colleagues is extremely important.’ While there is no data about the impact of LLMs on cardiovascular edu-

8 AI Emergency departments voice concerns Understaffed and overcrowded: survey highlights ED safety fears A survey of emergency medicine practitioners across Europe has highlighted major concerns over safety and overcrowding in Emergency Departments (ED). Understaffing, too few doctors and the risk of burnout were also among the issues high- lighted in the Europe-wide sur- vey by the European Society of Emergency Medicine (EUSEM). While AI technology may help al- leviate some of these issues, its implementation is connected to new challenges. Provisional findings of the survey, which is still in progress, were presented at the EUSEM 2023 con- gress in Barcelona by Dr Said Ha- chimi-Idrissi in a session which also focused on how technology can impact safety and quality with- in the ED. During his presentation, he admitted that the scenario was complex with no clear solutions. ‘The objective of the survey is to get an insight on Quality and Safety programme initiatives across Europe and more specifically in emergency departments and/or in Pre-Hospital EM settings,’ added Dr Hachimi-Idrissi, who is Clinical Chief of the Emergency Depart- ment (ED) at UZ Ghent in Belgium. More than a matter of mere discom- fort A key finding of the survey so far is that about 90% of professionals feel that at times the number of pa- tients in their ED exceeds the ca- pacity of the department to pro- that vide care safe and m o c . e b o d a . k c o t s - f f o k n e d o r o G overcrowding is a regular and seri- ous problem. Dr Hachimi-Idrissi added: ‘Overcrowding is not just an issue of discomfort or loss of dig- nity, but carries a substantial risk of harm and increased mortality. Understaffing was another major contributory factor.’ He said there are far too few doc- tors for the volume and complexity of many of the patients. This in- creases the risk of harm to the pa- tient, and of morbidity and mortal- ity, when the length of stay at the ED is too long. In addition, this im- balance is imposing stress, which may result in burnout, practitioners leaving the profession, impacting relationships and use of illicit sub- stances. The expert hopes the sur- vey findings help lead to changes in emergency care practice, in- crease awareness of the risk, and mitigate and tackle that risk to de- liver improvements. Adminis- trators, nurses, physicians, and pa- tients all have a role to play in achieving that, he added, naming development of clinical inter- ventions to reduce adverse events, better monitoring, the use of KPI (key performance indicators) and benchmarking as viable option to move towards this goal. AI: a new hope for the ED Technology could also have a role to play in making more efficient use of time in the ED and impact positively on quality and safety, such as use of Artificial Intelligence (AI) tools. ‘AI and machine learning would help to monitor patients and inte- grate data, vital signs and symp- toms to define the population at risk, though more research in this area is needed,’ Dr Hachimi-Idrissi said. However, he remains con- cerned that the current environ- ment in EDs across Europe is a health hazard for patients and healthcare providers. The session also included presen- tations looking at how hi-tech can catch impact positively on quality and safety in the ED; the use of big database and AI on continuous quality and safety improvement in emergency medicine; and how to integrate patient satisfaction in big databases improve perform- ances in the ED. to indicate *In a separate EUSEM session, find- ings were presented from a study that the AI chatbot ChatGPT performed as well as trained doctors in suggesting likely diagnoses for patients being as- sessed in the ED1. Said Hachimi-Idrissi Dr Said Hachimi-Idrissi is Clinical Chief of the Emergency Department at UZ Ghent in Belgium and Profes- sor of Critical Care Medicine at the University of Ghent as well as Profes- sor at the Free University Brussels. A Past President of the Belgian Society of Emergency and Disaster Medicine, he is Director of the Global Network on Emergency Medicine, a council member of the paediatric section of the EuSEM and has published exten- sively in peer-reviewed international publications. acknowledge While researchers at Jeroen Bosch Hospital, s-Hertogenbosch, The Netherlands that more work is needed, their find- ings suggest the technology could support doctors working in emerg- ency medicine, potentially leading to shorter waiting times for pa- tients. ■ Report: Mark Nicholls Real-time drug identification AI camera detects incorrect medication transfers With high proficiency, a deep- learning model identified con- tents of vials and syringes, con- firming whether medication transfers were correct. A team of researchers says it has devel- oped the first wearable camera system that, with the help of AI, detects potential errors in medi- cation delivery. In a new test, the video system recognized and identified, with high proficiency, which medications were being drawn in busy clinical settings. The AI achieved 99.6% sensi- tivity and 98.8% specificity at de- tecting vial-swap errors. The system could become a critical safeguard, especially in operating rooms, intensive-care units and emergency-medicine settings, said co-lead author Dr. Kelly Mi- chaelsen, an assistant professor of anesthesiology and pain medicine at the University of Washington School of Medicine. “The thought of being able to help patients in real time or to prevent a medi- cation error before it happens is very powerful,” she said. “One can hope for a 100% performance but even humans cannot achieve that. In a survey of more than 100 an- esthesia providers, the majority de- sired the system to be more than 95% accurate, which is a goal we achieved.” Drug administration errors are the most frequently reported critical incidents in anesthesia, and the most common cause of serious medical errors in intensive care. In the bigger picture, an estimated 5% to 10% of all drugs given are as- sociated with errors. Adverse events associated with injectable medications are estimated to affect 1.2 million patients annually at a cost of $5.1 billion. Syringe and vial-swap errors most often occur during intravenous in- jections in which a clinician must transfer the medication from vial to syringe to the patient. About 20% of mistakes are substitution errors in which the wrong vial is selected or a syringe is mislabeled. Another error. 20% of errors occur when the drug labeled is correctly but ad- ministered Safety in measures, such as a barcode sys- tem that quickly reads and con- firms a vial’s contents, are in place to guard against such accidents. But practitioners might sometimes forget this check during high-stress situations because it is an extra step in their workflow. The researchers’ aim was to build a deep-learning model that, paired with a GoPro camera, is sophis- ticated enough to recognize the contents of cylindrical vials and syringes, and to appropriately render a warning before the medi- cation enters the patient. Training the model took months. The inves- tigators collected 4K video of 418 drug draws by 13 anesthesiology providers rooms where setups and lighting varied. The video captured clinicians ma- naging vials and syringes of select medications. These video snippets were later logged and the contents of the syringes and vials denoted in operating to train the model to recognize the contents and containers. tecting a syringe that is lying on the table,” Gollakota said. The video system does not directly read the wording on each vial, but scans for other visual cues: vial and syringe size and shape, vial cap color, label print size. “It was par- ticularly challenging, because the person in the OR is holding a syr- inge and a vial, and you don’t see either of those objects completely. Some letters (on the syringe and vial) are covered by the hands. And the hands are moving fast. They are doing the job. They aren’t pos- ing for the camera,” said Shyam Gollakota, a coauthor of the paper and professor at the UW‘s Paul G. Allen School of Computer Science & Engineering. Further, the computational model had to be trained to focus on medi- cations in the foreground of the frame and to ignore vials and syr- inges in the background. “AI is doing all that: detecting the spe- cific syringe that the healthcare provider is picking up, and not de- This work shows that AI and deep learning have potential to improve safety and efficiency across a number of healthcare practices. Re- searchers are just beginning to probe the potential, Michaelsen said. included study also The re- searchers from Carnegie Mellon University and Makerere University in Uganda. The Toyota Research In- stitute built and tested the system. The Washington Research Foun- dation, Foundation for Anesthesia Education and Research, and a National Institutes of Health grant (K08GM153069) funded the work. The authors’ declared their poten- tial conflicts of interest in their paper, which will be made avail- able on request. ■ Source: University of Washington School of Medicine EUROPEAN HOSPITAL Vol 33 Issue 4/24

AI 9 Artificial intelligence meets internal medicine Medical AI: Enter ‘dea ex machina’ and to validate the trained models thoroughly. ‘We need to scrutinise generative AI very critically and de- velop good metrics.’ She concluded that the transparency of the models and explainability of the results must always be ensured. As an alternative to the current models, which are tailored to nar- rowly defined tasks, more recent research is moving in the direction of a “generalist” AI. This is de- signed to integrate any data moda- lities from the clinic and practice into an application and thus enable a holistic view of medical issues – such a model might at least come close to the concept of the deus ex machina. ‘Whether such a model will exist at all remains to be seen,’ said the ex- pert, pointing out the high level of complexity required. Instead of having a single “jack-of-all-trades” AI, she suggested that individual, specialised models might also be orchestrated with the help of a par- ent AI – a so-called LLM agent. De- pending on the clinical task, this supervising algorithm would “con- sult” the appropriate AI in order to generate a diagnostic or treatment recommendation based on its ex- pertise. The final hurdle that Wiest ad- dressed was the integration of the models into existing healthcare in- frastructures. ‘Interoperability of digital systems is urgently needed for this.’ This means there are still a few challenges to overcome before AI can make the leap from digital assistant to a fully-fledged clinical dea ex machina. ■ Report: Wolfgang Behrends ) e g a m i d e t a r e n e g I A ( m o c . e b o d a . k c o t s – a n n A © pathology, through organ seg- mentation to support surgical treat- ment planning, assessing radiological images. However, she emphasised that irrespective of the specialist area, it is fundamentally important to have a solid data basis to ACCUTRON® MR. THE MOBILE MRI SPECIALIST. Quick conﬁ guration, versatile use and compatible with up to 7 teslas. Accutron® MR allows a high patient throughput. Next, please. Hauptstrasse 255 · 66128 Saarbruecken www.medtron.com quantitatively, much manual re- work has been necessary up to now. Traditionally, tasks such as transferring the data from doctors‘ letters into a table have fallen to the doctoral students. Locally-run language models help protect sensitive data This tedious and unrewarding work could be delegated to a gen- erative AI, explained the expert, re- ferring to what is currently prob- ably the best-known application: The large language model (LLM) ChatGPT from OpenAI. This pro- gram has been trained on large text corpora and is able to generate new content on this basis. This also includes extracting information from unstructured texts. One re- cent study has demonstrated useful results in retrieving TNM stages (tumour/node/metastasis) from pathological reports – even if these had not been digitised. However, the OpenAI solution can- not be used in these cases, as the sensitive patient data should never be transferred to the US company‘s servers, Wiest pointed out. A viable alternative would be locally-run language models such as Llama-2, which can be installed on a com- puter on site. ‘This means that the data stays where it is generated and does not leave the hospital or practice.’ Despite having far less computing power, the local AI model delivered good results in tests with gastroenterological en- doscopy findings, the expert re- ported. AI-generated summaries free up mo- re time for patient care that contain Initial studies suggest that the LLMs also perform well when sum- marising the content of clinical texts. Although these summaries are by no means flawless, Wiest re- the AI-generated cognised texts fewer misinter- pretations and inaccuracies than those of human experts. The ex- pert argued that the use of these tools could ‘make work easier and increase efficiency, leaving more time for the patient and less time for documentation’. However, for LLMs to provide medically accurate information without producing the infamous “hallucinations”, the input has to be right, Wiest pointed out: for example, current guidelines would have to be entered into a database in the form of machine-readable vectors so that the corresponding prompt results in well-founded answers. This approach, known as ‘Retrieval Augmented Generation’ (RAG), has already produced some good results in initial trials, accord- ing to the expert. ‘This is a promis- ing technology that doctors can use to have up-to-date treatment recommendations available when they are needed.’ One algorithm for all – or all for one The increasing digitalisation of health data opens up numerous other potential applications for AI, Wiest explained: From predicting microsatellite instability (MSI) to detect genetic changes in histo- In the world of theatre, the ‘deus ex machina’, the god from the machine, is a dramaturgical trick to resolve seemingly unsolvable conflicts. Can artificial intelli- gence (AI) also be such a univer- sal problem solver for internal medicine At the Annual Con- gress of the German Society of Internal Medicine (DGIM) in Wiesbaden, Dr Isabella Wiest ex- plored the potential – and limi- tations – of AI helpers. The research associate at the Else Kröner Fresenius Centre (EKFZ) for Digital Health at TU Dresden and resident physician at Mann- heim University Hospital explained that the aspect of generative AI in particular shows enormous poten- tial. Every person generates a large quantity of health-related data on their way through doctors‘ sur- geries and clinics, but also in their private lives, and this data could be analysed by an AI for clinically rel- evant patterns, the expert pointed out. Examples include medical re- ports, radiology images, histopath- ology scans as well as genome and sensor data. However, the availability of this wealth of data is only half the battle, Wiest pointed out: around 80% of data is stored in un- structured form, and the free text of findings often contains relevant additional information that goes beyond the structured data. In order information to use this www.healthcare-in-europe.com

10 Transformative technology Debating the case for AI in pathology AI ‘The integration of AI in pathology has been limited by a range of ob- stacles: the slow diffusion of digital pathology, which is the basis of AI, uncertainties about the quality of AI algorithms and the regulatory constraints,’ she said. Elements missing for large-scale adoption of AI in real life settings include seamless integration of algorithms in the pathologist workflow, a sus- tainable economical model and lack of an in-depth reflection on the strategy for using AI within the discipline. However, in a survey during the session, more than half of dele- gates found themselves supporting the “pro” side, agreeing, or even strongly agreeing, that pathologists are ready for using AI in practice. How much do we need to understand A second debate focused on inter- pretability and explainability in AI and whether it is necessary. Vin- cenzo L‘Imperio, Assistant Pro- fessor of Pathology at the Univer- sity of Milano-Bicocca, Monza, Italy, argued that it is. ‘Inter- pretability and explainability in AI can be of help when applying these algorithms for diagnostic purposes,’ he said. However, Diana Montezuma Felizardo the clinical laboratory group IMP Diag- nostics in Portugal, argued against the question. She said: ‘AI has made significant advances in vari- ous fields, including pathology. As AI models become more complex, the topics of interpretability and explainability have risen as key concerns. The real issue often lies in the delicate balance between making AI understandable while maintaining its sophisticated capa- bilities. ■ from Report: Mark Nicholls Artificial Intelligence (AI) re- mains a divisive topic within the discipline of pathology with a range of opinions over its cur- rent value and applicability in clinical settings. While most ex- perts agree that the technology will not replace pathologists, it might still spell bad news for those who do not embrace AI in their daily practice. On the other hand, reservations persist about whether pathologists are ready to achieve this adoption this in the first place. The topic was discussed during a lengthy session at the 36th Euro- pean Congress of Pathology in Flo- rence, Italy, where speakers argued the pros and cons of AI within pa- thology in a lively debate. Dele- gates were also asked to add their views during real-time surveys on pertinent questions regarding AI. Will Europe be left behind Yet it was a powerful opening from a leading AI expert that set the tone for the session. Dr Ricardo Baptiste Leite, CEO of HealthAI – a Geneva-based global agency for re- sponsible AI in health – said the global AI healthcare market was worth $11bn in 2022, with China taking a lead with 61% of patents in 2015–22, compared to 21% from America and a mere 2% from the UK and the EU. The market worth is estimated to reach nearly half a trillion US Dollars in 2032. ‘We are on the verge of tremen- dous geopolitical tensions in which Europe really needs to reposition itself if it wants to be an active player,’ he said. Leite indicated that AI may be accelerating healthcare inequalities and that it was import- ant to bring the human factor into the conversation and ensure that the ‘use of technology can lead to better health outcomes’ for pa- tients. Navigating the “wild west of AI” The expert pointed out key AI trends, including healthcare ana- lytics, medical diagnostics, tele- health, robotics, management and clinical trials. On the other hand, significant challenges remain, such as cybersecurity risks, regulatory and ethical issues. There is also global fragmentation with different requirements, approval processes and timelines across different countries. While AI shows clear potential to save lives, Leite underlined the need to get regulation right. ‘Lack of effective governance of AI in health increases risks for citizens, and we are living in an age that I would define as the “wild west of AI”. In many contexts, people are using this technology with no over- sight, and companies are putting this technology in countries where there is no data protection.’ As a former Parliamentarian, he there- YOUR ADVERTISEMENT OR ADVERTORIAL Please cnac ur media cnsulan Toni Lauterbach Tel.: +49 (0)9221 949374 t.lauterbach@mgo-fachverlage.de m o c . e b o d a . k c o t s - o t y c r a i © fore advocates ‘light legislation, ef- fective regulation’ for AI and ex- pressed his belief that politicians relinquish power on the topic more to regulatory bodies. Prediction: AI will change pathology – and pathologists As the session moved on, the panel debated whether pathologists are even ready for AI in practice. features, Arguing in favour, Prof. Dr Falko Fend from the Institute of Pathol- ogy at Tübingen University Hospi- tal in Germany, discussed what pathologists want from AI, such as increased diagnostic accuracy and extraction of “invisible” biologic and prognostic and looked at the AI tools that are needed in pathology. But he warned: ‘To use it in a technical way, we need an integrated digital pathology system; we need ex- pertly curated cohorts, clinically and diagnostically valid questions, intensive collaboration with com- puter scientists, stringent evalu- ations in practice settings, and data.’ Bottlenecks to AI are the availabil- ity of digital pathology, computing power and storage, and a need for regulation. While there are doubts over jobs and skills erosion, the ex- pert said there is a need to foster a ‘pragmatic attitude’ to AI. ‘Pathol- ogists need to be involved right from the beginning, testing al- gorithms in clinical diagnostic setting,’ he continued. ‘In the end, AI should not replace expertise in pathology, but pathol- ogists should use AI tools to make a better diagnosis and not see AI as a replacement for expertise. Per- haps AI will make us better pathol- ogists, but different ones to before.’ rigorously Exploratory phase Providing the con counterpart to Prof Fend’s pro-arguments, Dr Ca- therine Guettier from the Depart- ment of Pathology, Hôpital Bicêtre, Paris, suggested pathologists are not ready for AI in practice. She de- clared that AI in pathology is still in the exploratory phase, with only very few early adopters. To back up this claim, the expert pointed out that, as of April 2023, there were fewer than ten AI-enabled medical devices in pathology ap- proved by the FDA – compared to nearly 400 in radiology. Falko Fend Professor Dr Falko Fend is Medical Director at the Institute of Pathology and Neuropathology and Head of the Department of General and Molecu- lar Pathology and Pathological Anat- omy at Tübingen University Hospital in Germany. Catherine Guettier Catherine Guettier is Professor of Pathology at Paris-Saclay University and Head of the Department of Pa- thology Hôpitaux Universitaires Paris-Saclay – Assistance publique – Hôpitaux de Paris (AP-HP). A pioneer of digital pathology in France, her ex- pertise is in liver pathology and de- velopment of AI-based algorithms in pathology. Ricardo Baptista Leite Dr Ricardo Baptista Leite is a Por- tuguese-Canadian medical doctor trained in infectious diseases with extensive experience in global health. Prior to his current role as CEO of HealthAI, he served four terms as a Member of Parliament in Portugal. Dr Diana Montezuma Felizardo is Head of R&D at IMP Diagnostics, a group of clinical diagnostic labora- tories specializing in anatomic pa- thology and headquartered in Porto, Portugal. Vincenzo L‘Imperio is Assistant Professor of Pathology at the Univer- sity of Milano-Bicocca, Monza, Italy. His professional interests focus on renal pathology, hematopathology and proteomics. EUROPEAN HOSPITAL Vol 33 Issue 4/24

RESEARCH 11 Hyperbaric oxygen therapy to reduce toxic effects HBOT shows promise after breast cancer radiotherapy oxygen radiotherapy Hyperbaric therapy (HBOT) treatments could offer relief to breast cancer patients who experience late toxicities following treat- ment. To date, the handful of completed clinical trials only produced inconclusive or contra- dictory results. Therefore, results from the latest trial, named HONEY (Effect of Hyperbaric Oxygen Therapy on Breast Cancer Patients with Late Radi- ation Toxicity), were anxiously awaited. However, re- searchers’ insights, published in JAMA Oncology, also do not pro- vide much definitive evidence of HBOT’s benefit.1,2 the conditions, HBOT provides 100% oxygen for inhalation inside a pressurized treatment chamber. The therapy is used to treat a large variety of medical including toxicities caused by cancer treat- ments, by increasing the amount of oxygen that the blood can carry throughout the body and by de- livering oxygen-rich plasma to hy- poxic can help strengthen immune responsiveness and encourage new collagen de- position and endothelial cell formation.3 tissue. This The HONEY trial A multi-institutional team in the Netherlands conducted the hospi- tal-based randomised clinical trial to assess the effectiveness of HBOT to reduce or eliminate moderate or severe breast, chest wall, and/or Brain cancer immunotherapy shoulder pain in combination with edema, fibrosis, or movement re- striction 12 months or longer fol- lowing breast irradiation. Three hundred eighty-eight (19.1%) pa- tients out of 2,029 survey re- sponders (who had been partici- pants in a previous breast cancer clinical trial) reported they were experiencing these late toxic ef- fects. A total of 189 were eligible to participate in the HONEY trial. invited researchers 125 The women to receive 30 to 40 HBOT sessions 120 minutes in duration over a period of six to eight con- secutive weeks. Eighty-five women declined the treatment, with 70 saying that the time commitment required to participate was too great. An additional 61 were se- lected as the control group. Ultimately, only 31 women com- pleted the HBOT treatment, which consisted of breathing 100% oxygen through an oxygen mask during four intervals of 20 minutes at an atmospheric pressure of 2.5 ATA (or bar). Principal investigator Prof. Helena M Verkooijen, MD, PhD and colleagues evaluated these patients at baseline and at three- and six-months post-treat- ment. They report that in follow up, moderate to severe pain was not statistically different between the HBOT group (50%) and the control group (62%). However, the HBOT treatment participants ex- perienced a significant reduction in fibrosis. In an accompanying editorial, Dr Ezra Hahn of the Princess Margaret Cancer Centre in Toronto, Prof. Dr Aron Popovtzer of the Hadassah University Medical Center in Jeru- salem, and Prof. Dr Benjamin W. Corn of the Hebrew University Fac- ulty of Medicine in Jerusalem, write: ‘We now have modest evi- dence to justify the use of HBOT in treating the chronic morbidities as- sociated with breast irradiation. […] In terms of the next steps, ad- equately powered randomized, sham-controlled, double-blind trials will be necessary to truly de- termine the benefit of HBOT.’ 2 reduce Breast cancer-related HBOT clinical trials in progress Two clinical trials are currently underway in Asia to evaluate how HBOT may toxicities caused by radiotherapy to treat breast cancer. South Korean re- searchers at the Cha Ilsan Medical Center, a woman’s hospital in Goy- ang-si near Seoul, are conducting a clinical trial to determine the effi- cacy and safety of HBOT to pre- vent radiation dermatitis. Patients receiving adjuvant radiotherapy following breast conservation sur- gery for localized invasive breast cancer are either randomized to re- ceive HBOT three times a week after completing five fractional radiotherapy treatments (for 60 Gy total) or to receive standard edu- cation in dermal care. Principal in- vestigator Assistant Professor Jee Young Lee advises that the study, which started November 2023, is measuring the incidence and sever- ity of radiation dermatitis in each group, as well as evaluate any self- described quality of im- provements. 4 life A large randomized clinical trial of nearly 400 breast cancer patients receiving treatment at Guangdong Provincial People’s Hospital in Guangzhou, China, is evaluating if HBOT treatments can prevent radi- ation pneumonitis caused by chemotherapy and radiotherapy. Radiotherapy can cause radiation lung injury, leading to pulmonary fibrosis. 5 the secretion of thereby ‘Hyperbaric oxygen can improve the tissue after radiation by pro- moting the function of vascular en- dothelial cells and fibroblasts, and reducing in- flammatory factors, in- hibiting the process of fibrosis and fiber atrophy after radiotherapy, and promoting tissue repair. There- fore, it has the potential value of treating chronic radiation injury,’ explains principal investigator Prof. Kun Wang, MD, PhD. therapy Half of the patients are receiving standard-of-care treatment. The other half are receiving 30–40 hy- perbaric oxygen treat- ments immediately after the end of radiotherapy. Researchers will evaluate the incidence of radiation pneumonitis 12 months following completion of treatment for both patient cohorts, and the type and severity of symptoms among both groups of patients who develop radiation pneumonitis.■ Article: Cynthia E. Keen References 1 Moss AJ et al.: Cardiac-Resynchronization Therapy for the Prevention of Heart-Failure Events; New England Journal of Medicine 2009; https://doi.org/10.1056/ NEJMoa0906431 2 Bristow MR et al.: Cardiac-resyn- chronization therapy with or without an im- plantable defibrillator in advanced chronic heart failure; New England Journal of Medi- cine 2004; https://doi.org/10.1056/ nejmoa032423 3 Cleland JGF et al.: The effect of cardiac re- synchronization on morbidity and mortality in heart failure; New England Journal of Medicine https://doi.org/10.1056/ nejmoa050496 2005; 4 Tang ASL et al.: Cardiac-resynchronization therapy for mild-to-moderate heart failure; New England Journal of Medicine 2010; https://doi.org/10.1056/nejmoa1009540 5 Maytin M, Jones SO, Epstein LM: Long-term mortality after transvenous lead extraction; Circulation: Arrhythmia Electro- physiology 2012; https://doi.org/10.1161/ circep.111.965277 and Turning immune cells against glioblastomas Glioblastoma is the most com- mon and most aggressive pri- mary brain tumour, with an aver- age survival after diagnosis of less than two years, and against which current treatments remain ineffective. carry biological Glioblastomas characteristics that make them par- ticularly difficult to treat. Able to induce a microenvironment that li- mits the attack of the immune sys- tem, they escape standard treat- ments and recur rapidly. In recent years, immunotherapies have given patients renewed hope, albeit with relatively modest suc- cess. A team from the University of Geneva (UNIGE) and the Geneva University Hospitals (HUG) has succeeded in identifying a specific marker on the surface of tumour cells, and in generating immune cells carrying an antibody to de- stroy them. Furthermore, these cells, called CAR-T cells, appear to be capable of targeting diseased cells in the tumour that do not carry this antigen, while sparing healthy cells. These results, pub- lished in the journal Cancer Immu- nology Research, are a first step to- wards the development of clinical trials with human patients. the Denis Migliorini, assistant pro- fessor in the Department of Medi- cine at the UNIGE Faculty of Medi- cine, holder of ISREC Foundation Chair in Brain Tumour Immunology, member of the Trans- lational Research Centre in Onco- Haematology (CRTOH) and attend- ing physician in charge of the HUG Neuro-oncology Unit, is an expert in CAR-T cells (for chimeric antigen receptors T-cells). This im- munotherapy consists in collecting immune T cells from patients, mod- ifying them genetically in the lab to make them express antibodies ca- pable of detecting elements spe- cific to tumour cells, before re- injecting them so that they can specifically target the tumour. www.healthcare-in-europe.com to identify ‘‘For several years we have been trying the protein markers expressed by glioblastoma cells,’’ explains Denis Migliorini. ‘‘One of these markers, PTPRZ1, proved particularly important: we were able to generate CAR-T cells carrying targeting PTPRZ1. This is a first step towards CAR-T cells effective against malig- nant gliomas.” antibodies Most CAR-T cells are generated using viral vectors, a technique that has proved its worth in certain dis- eases but is not very suitable in the brain. ‘‘Indeed, they persist for a very long time in the context blood cancers. The brain is a fragile organ, and this persistence can generate a risk of toxicity,’’ ex- plains Darel Martinez Bedoya, a post-doctoral fellow in Denis Mig- liorini’s laboratory and first author of this research. The scientists therefore introduced in the T-cells the messenger RNA encoding for the desired antibody. The cellular machinery is then responsible for The second stage involved testing the treatment in vivo in mouse models of human glioblastoma. Tu- mour growth was controlled, pro- longing the lives of the mice re- markably well without signs of toxicity. ‘‘By administering CAR-Ts intratumourally in the CNS, we can use fewer cells and greatly reduce the risk of peripheral toxicity. With this data and other unpublished yet, all lights are green to now en- visage a first clinical trial in hu- mans,’’ the scientists conclude. ■ Source: University of Geneva producing the right protein to build the receptor that will take place on the T-cell surface and re- cognise the tumour target. ‘‘This technique has a number of ad- vantages: CAR-Ts offer a flexible platform, allowing multiple adap- tations according to the specifi- cities and evolution of the tumour,’’ explains Darel Martinez Bedoya. To check that CAR-Ts only attack tumour cells, the Geneva team first tested them in vitro on healthy and tumour cells. ‘‘To our surprise, not only did CAR-Ts not attack healthy cells, but they were also capable, by bystander effect, of identifying and fighting tumour cells not ex- pressing the PTPRZ1 marker,’’ Denis Migliorini is delighted to re- port. ‘‘In this context, CAR-Ts are probably capable of secreting pro- inflammatory molecules that are responsible for eliminating tumour cells even in the absence of the original marker when co-cultured with target positive tumour cells.’’

12 RESEARCH Experts explore the future of CSP, CRT, ICD Implantable cardiac devices: which techniques are ready for prime time Opposing views on new implant- able cardiac devices were aired in a Great Debate session at the European Society of Cardiology’s annual 2024 congress in Lon- don. Experts discussed emerging techniques and technologies and debated whether they are ac- tually ready for clinical ap- plication. At the core of the session was the issue of whether conduction system pacing (CSP) should replace cardiac resyn- chronisation. Speakers firstly discussed the pros and cons of whether CSP – a car- diac pacing technique that uses the heart’s natural conduction system to help the ventricles activate in a physiological way via leads in the organ’s – should resyn- chronisation, a treatment for heart failure (HF) im- planting a device to help the heart’s ventricles pump in a coordi- nated way. conduction replace cardiac system that involves A second element to the Great De- bate focused on whether patients with an ICD (implantable cardio- verter-defibrillator) but no indi- cation for bradycardia pacing should receive an extravascular ICD (EV-ICD), where the lead is placed below the breastbone and is outside of the heart. CSP: A proximal solution for a proximal problem Arguing the case for CSP to replace cardiac resynchronisation therapy (CRT), Dr Jacqueline Joza from McGill University Health Centre in Montreal in Canada, felt the debate should actually focus on whether CSP should replace biventricular (BV) pacing as both methods re- synchronise the heart. Her key points were to suggest that the HIS-Purkinje system was meant to conduct rapidly and synchronously (so it should be used); that the block in LBBB (left bundle branch block) occurs most often at a lo- cation within the HIS (the cardiac muscle cells that transmit electrical impulses), so it is a case of deliver- ing a proximal solution for a proxi- mal problem; and that BV-CRT is an inefficient and dyssynchronous method of delivering CRT. Joza conceded: ‘The only time BV- CRT may be a better fit is when there is no HIS-Purkinje disruption at all.’ But she added that CSP per- forms well in patients with RBBB (right bundle branch block) and HF indication for pacing. Just a ‘new kid on the block’ in need of evidence Putting the case against CSP re- placing CRT was Cardiologist- Electrophysiologist Professor Kevin Vernooy from Maastricht University Medical Centre in the Netherlands. He argued that there are many trials that show BV pacing is much better than just an ICD in these pa- tients1 with much evidence to sug- gest that BV in patients with LBBB and HF is an ‘awesome therapy.’2,3,4 While acknowledging CSP in HF is a ‘new kid on the block’, he ex- pressed surprise that clinicians seem to be moving towards it ‘without evidence’ when he be- lieves there is such a good existing pacing therapy. ‘We have so much evidence that biventricular pacing is doing a good job,’ added Ver- nooy. ‘Even though CSP is theor- etically an alternative for BV pac- ing, the implantation requires new skills and knowledge with limited success rates in patients with heart failure with tools. Therefore, we still need BV as a bail-out strategy, so I think we should not go too quickly.’ the current He concluded that while CPS is a ‘promising’ pacing strategy, clinical routine is shifting towards it ‘with- out evidence’; that training, edu- cation and EP understanding is ‘es- sential’; and that CSP does not seem to be the ‘optimal strategy’ in many patients. Transvenous leads: only for those that need them In a second debate in the session, Professor Lucas Boersma from St Antonius Hospital in Nieuwegein, the Netherlands, argued the case for patients with an ICD but no in- dication for bradycardia pacing, re- ceiving an extravascular ICD (EV- ICD). While acknowledging that ICDs save lives and the Transvenous ICD (TV-ICD) has become the ‘work- horse’ over the last 3–4 decades in everyday clinical care for Sudden Cardiac Death (SCD), he said TV- ICD therapy comes with a price of morbidity and mortality and that leads can become infected or break down.5 ‘We used to take that for granted because there was no al- ternative,’ he said. ‘But if we place the lead outside the vasculature, that may avoid unnecessary com- plications.’ S-ICD (subcutaneous) was an im- portant first step but he pointed to further advantages of EV-ICD: smaller devices, lower defibrillator energy, greater longevity, pacing with a single device, anti-tachyc- ardia pacing, asystole pacing sup- port and post-shock-pacing. Boers- ma concluded by pointing out that EV-ICD and S-ICD have high con- version rates and similar efficacy for transforming ventricular ar- rhythmias back to sinus rhythms and they avoid harmful vascular complications. ‘For patients that do not need pacing, these devices with no transvenous lead in the heart are the way to the future; we should not give transvenous leads to patients that don’t need them.’ TV-ICD: ‘still a lot to offer’ Offering the opposing view was Professor Jeanne Poole from the University of Washington in Seattle. Her institution is a high user of S-ICD, which has been FDA approved for 14 years, so she sug- gested it is already ‘prime time’ in appropriate patient groups, where- as EV-ICD has only been recently approved in the United States and has not been adequately tested in real world prospective registries or randomised controlled trials. ‘So, categorically, it is not ready for prime time,’ she said. fracture, The expert discussed the risk of lead though acknowl- edged future transvenous leads are likely to be more robust and smaller, and that infection rates are now significantly lower. Pointing to additional considerations in favour of TV-ICD systems, she said they still have the lowest inappropriate shock rates and set a ‘high bar’ with tens of thousands of patients enrolled in clinical studies. There is also excellent sensing with endoc- ardial leads, expanded diagnostics, long battery life, well-established surgical techniques, no need for general anaesthesia, painless ATP for patients, and offer flexibility in programming. In summary, Prof. Poole said that all patients with an ICD indication but no indication for bradycardia pacing should categorically not re- ceive an EV-ICD. ‘Transvenous ICD systems still have a lot to offer, they are the only option for many ICD indicated patients and they will be around for many years,’ she concluded.■ Lucas Boersma Professor Lucas Boersma is an Electrophysiologist at St Antonius Hospital in the Netherlands and was Head of the Cardiology Department from 2008–2016. He is the past chairman of the Committee for Science & Innovation of the Nether- lands Society of Cardiology and en- dowed Professor of Cardiology at the University of Amsterdam since 2017. Jeanne Poole Professor Jeanne Poole is an elec- trophysiologist and the Section Head of Electrophysiology at the University of Washington in Seattle, US, and is Editor-in-Chief for the Heart Rhythm O2 Journal. Jacqueline Joza Dr Jacqueline Joza is an associate professor of medicine and cardiac electrophysiologist at the McGill Uni- versity Health Centre, Canada. Her research focuses on CSP, atrial fibril- lation and syncope. Kevin Vernooy Professor Kevin Vernooy is the medical director of the heart and vascular center and chairs the De- partment of Cardiology at the Maas- tricht University Medical Center in the Netherlands. His research focuses on the invasive treatment of cardiac arrhythmias in patients with heart failure. Publisher mgo fachverlage GmbH & Co. KG E.-C.-Baumann-Str. 5 95326 Kulmbach/Germany Phone +49 (0) 9221 949-311 Fax +49 (0) 9221 949-377 Managing Directors: Eva-Maria Bauch, Stephan Behrens Editor and Production Manager: Tim Hofmann (TH) Phone: +49 (0) 9221 949-176 E-Mail: redaktion@european-hospital.com Editor: Wolfgang Behrends (WB) Founded by Heinz-Jürgen Witzke ISSN 0942-9085 Correspondents Austria: Michael Krassnitzer (MK) Germany: Cornelia Wels-Maug (CWM) Dr Christina Czeschik (CC) Great Britain: Mark Nicholls (MN) Spain: Mélisande Rouger (MR) USA: Cynthia E. Keen (CEK) Subscriptions Lena Krauß kundenservice@mgo-fachverlage.de Subscription rate 4 issues: 32 Euro, Single copy: 8 Euro. Printed by: mgo360 GmbH & Co. KG, Bamberg, Germany Publication frequency: quarterly Representatives Germany, Austria, Switzerland: Toni Lauterbach Phone: +49 9221 949-374 E-Mail: t.lauterbach@mgo-fachverlage.de France, Italy, Spain: Eric Jund Phone: +33 493 58 77 43 E-Mail: jund@european-hospital.com GB, Scandinavia, BeNeLux: Simon Kramer Phone: +31 180 6200 20 E-Mail: kramer@european-hospital.com Taiwan: Charles Yang Phone: +886 4 232 236 33 E-Mail: medianet@ms13.hinet.net USA & Canada: Hanna Politis, Media International Phone: +1 301 869 66 10 E-Mail: hanna@media-intl.com All company, brand and product names in this publication are the property of their respective holders. Users must obtain per- mission from those holders before copying or using the owner’s trademarks, product and company names or logos. EUROPEAN HOSPITAL Vol 33 Issue 4/24