2 PAT H O L O G Y / L A B O R AT O R Y AI provides prognostic information Next-generation deep learning models predict cancer survival m o c . e b o d a . k c o t s – ) E & H ( z d o L y t i s r e v i n U l a c i d e M ; ) h p a r G ( i e h r a i S © Image source: Jiang et al., Lancet Digital Health 2022 (CC BY 4.0) Deaths from cancer are current- ly estimated at 10 million each year worldwide. Conventional cancer staging systems aim to categorize patients into different groups with distinct outcomes. ‘However, even within a specific stage, there is often substantial variation in patient outcomes,’ Markus Plass, academic research- er from the Medical University of Graz, Austria, explained to European Hospital. Hence the rapid growth of Artificial Intelligence (AI), machine learn- ing and deep learning in provid- ing novel prognostic information that is not captured in current staging guidelines. Deep learning, a subdivision of machine learning, uses convolu- tional neural networks to devise informative representations of raw input data automatically, without requiring manual feature engineer- ing. ‘This is particularly useful for image segmentation and classifica- tion in histology slides,’ said Athena Davri, biologist at the Department of Pathology in the Faculty of Medicine, School of Health Sciences, University of Ioannina in Greece. Currently, histopatho- logy examination of tissue remains the ‘gold standard’ for diagnosing colorectal cancer, the second most common cancer in women and the third most common in men. However, routine pathology lab tests are taking up a lot of time and effort, due to the high incidence of this type of cancer. Furthermore, the worldwide shortage of patholo- gists has led to delays in diagnosis. AI models initially made it possible to automate and speed up the work before integrating parameters from the tumour ecosystem. So much so in fact, that in 2022 alone, Davri's team has listed, in a systematic review published in August, around a hundred scien- tific articles devoted to deep learning on histopathology ima- ges for diagnosing colorectal can- cer. According to this systematic review, ‘algorithms based on deep learning have the potential to assist with diagnosis, identify histolog- ical features relating to progno- sis and associated with metastasis, and assess the specific components in the tumour microenvironment,’ said Davri. Contextual histopathology features from whole-slide images Today, the revolution in progno- sis is coming from the relation- ship between deep learning and whole slide imaging. Also known as virtual microscopy, whole slide imaging refers to scanning a com- plete microscope slide and creat- ing a single high-resolution digital file. A typical whole slide image may contain 100,000×100,000 pix- els. Analysing and viewing whole slide images is often constrained by computer memory or screen size. A common workaround for these issues is capturing many smaller, high-resolution image tiles or strips, which are then stitched together to create a full image of a single histological section. This works because whole slide scan- ners take separate images of each field of view across the entire slide at high speed. The images acquired separately are then stitched togeth- er during the scanning process to generate a single digital image at full resolution. Many experts agree that machine learning is the future for digital pathology. A Chinese team from the Sun Yat-sen University Cancer Center, Guangzhou, affiliated to the laboratory of Precision Medicine for Gastrointestinal Tumors at Nanfang Hospital, and US scientists from the Department of Radiation Oncology at Stanford University School of Medicine in California, have designed a multitask deep learning platform for simultaneous- ly predicting peritoneal recurrence and disease-free survival using pre- operative CT images. ‘We trained it using a retrospective, multi-in- stitution study on 2,320 subjects and evaluated the prognostic accu- racy of the model as well as its association with chemotherapy response,’ said Prof Guoxin Li from the Department of General Surgery & Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Tumor. Findings were published in The Lancet Digital Health in May 2022. When informed by the AI model, clinician performance was signif- Pediatric Urology, the Department of Diagnostic and Interventional Radiology, and the Institute of Pathology all at the University Medical Center in Mainz, Germany, did develop and evaluate a mul- timodal deep learning model for prognosis prediction in clear-cell renal cell carcinoma (ccRCC). This disease is the most common type of kidney cancer with more than 175,000 associated deaths each year. In contrast to other tumour types, there is no clearly defined set of biomarkers used in clini- cal routine. Clinical management of ccRCC usually involves various disciplines including urology, radi- ology, oncology, pathology, and more besides. ‘This results in a vast amount of medical data on each patient, such as CT/MRI scans, histopathology images and other clinical information,’ explained Prof Axel Haferkamp, Director of the Department of Urology and Pediatric Urology. There are several clinical tools for prognosis prediction in ccRCC, such as the UCLA Integrated Staging System (UISS) or the International Metastatic Renal Cell Carcinoma Database Consortium risk model. ‘But while prognostic clinical nomo- grams might be helpful, they can be cumbersome to use and often only incorporate a selection of the available information – both of which potentially limit their per- formance,’ said Peter Mildenberger, radiologist, consultant and associate professor at the Department of Diagnostic and Interventional Radiology in Mainz. Their multimodal deep learning model showed great performance in predicting the prognosis of clear- cell renal cell carcinoma patients, with a mean accuracy of 83.43%. Furthermore, this latest tool’s pre- diction was an independent prog- nostic factor which outperformed other clinical parameters. senior Report: Bernard Banga icantly enhanced for predicting peritoneal recurrence. Additionally, the AI was able to identify which patients with stage II and stage III gastric cancer were most likely to benefit from adjuvant chemo- therapy. Multimodal deep learning close to clinical routine All forms of cancer stand to ben- efit from the prediction capabili- ties of machine learning models. One team, from Seoul National University in the Republic of Korea, is applying the machine learning approach to the analysis of whole- slide images of kidney, breast, lung and uterine cancers. In August, the researchers tested their deep learn- ing graph on 3,950 patients with these four types of cancer. ‘Deep graph neural networks that derive contextual histopathology features from whole slide images may aid diagnostic and prognostic tasks,’ said Prof. Kyung Chul Moon from the Department of Pathology at Seoul National University College of Medicine. Currently, no AI-based models are being used in clinical practice. However, multimodal approach- es linking biology, histopathology imaging and gene expression are bringing experimental deep learn- ing models closer to routine clini- cal application. In 2021, teams from the Department of Urology and Deep learning-based algorithm for Papillary thyroid carcinoma to According the World Health Organization, the tall cell variant (TCV) is an aggressive subtype of pap- illary thyroid carcinoma (PTC) featur- ing at least 30% epithelial cells two to three times longer than they are wide. In practice, applying this distinction is difficult, leading to substantial varia- tions between observers. ‘That’s why we are developing and training a deep learning algorithm using supervised learning to detect and quantify the proportion of tall cells (TCs) in pap- illary thyroid carcinoma,’ explained Sebastian Stenman, researcher from the Institute for Molecular Medicine, and the Department of Pathology at the University of Helsinki, Finland. In summer 2022, his research team tested it on an independent data set, and further validating it on an independent set of 90 papillary thy- roid carcinoma samples from patients treated at the Hospital District of Helsinki and Uusimaa (HUS) between 2003 and 2013. The Finnish scientist compared the algorithm-based tall cell ratio to independent scoring by a human investigator and looked at how those scores were associated with disease outcomes. The results published revealed that the deep learning algorithm detected tall cells with a sensitivity of 93.7% and a specificity of 94.5%. In the validation set, the deep learning algorithm tall cell scores correlated with a diminished relapse-free survival. ‘We showed that the DL-based algorithm was better than the human observer in identifying tall cell variants,’ said Stenmann. The algorithm could prove useful as a clini- cal tool for pathologists when evaluat- ing PTC samples and could potentially significantly improve the consistency of TCV case assessment. So far, no such algorithm has been described. EUROPEAN HOSPITAL Vol 31 Issue 4/22

4 PAT H O L O G Y / L A B O R AT O R Y m o c . e b o d a . k c o t s – i o d u t S d o o G © therefore be used,’ said S c h l a p p i . So, there is need for strin- gent optimization of reaction conditions and parameters in order to achieve a multiplex process. Multiplex Diagnostics in Research Although there are currently around a dozen companies as key players in multiplex diagnostics, there are many multiplex immuno- assays (MIAs) under development and only a few have so far been commercialized. One team, from the Department of Biomedical Engineering at the University of California, has demonstrated an approach using a smartphone ca- mera for reading ELISA-on-a-chip assays. This team designed a cell phone-based handheld microplate reader that uses optical fibres to transmit data from ELISA plates to a cell phone camera for diagnos- tics at the point of care. Similarly, researchers from Kyung Hee University, Yongin, South Korea, have been looking into an innovative multiplexing method for nucleic acid testing. In 2022, they published in Biosensors and Bioelectronics details of a rapid multiplexed molecular diagnostic system, dubbed a flow genetic ana- lysis system (FGAS), capable of con- ducting quantitative detection of nucleic acids. FGAS is portable and bat- tery powered, making it suitable for low resource settings. It connects to a smartphone, used for fluorescent imaging. This new system represents a fully automa- ted multiplex molecular diagnostic device for respiratory tract infec- tions. The platform is based around reverse transcriptase polymerase chain reactions (RT-PCR) and is capable of automated sample-to-an- swer analysis with a turnaround time of 3 hours 20 minutes, using centrifugal microfluidics. ‘Our diagnostic device, based on the Internet of Things (IoT), is accom- panied by an integrated micro- fluidic chip capable of running a multiplexed reverse-transcriptase LAMP,’ said Huynh Quoc Nguyen from the university’s Department of Chemical Engineering. Future lab-on-a-chip and lab-on-a disc platforms The development of micro- and nanofluidics has inspired the emer- gence of several miniaturized plat- forms, such as lab-on-a-chip and lab-on-a-disk. ‘These platforms deliver the capabilities of molecu- Clustered lar-scale sen- sitivity on low-cost and rap- idly fabricated devices,’ said Travis Schlappi. However, adopting these platforms into clinical diagnostics is yet to happen. Droplet microfluidics and microarray are some of the techniques used to achieve multiplexing using digital PCR (dPCR). While very promising, the development and commercialization of microfluidic platforms are held back by the high cost and complexity of manu- facturing on a large scale. Could Regularly Interspaced Short Palindromic Repeats (CRISPR) offer a solu- tion? In recent years, a number of studies have migrated towards the application of CRISPR/Cas12a systems for multiplex molecular diagnostics. CRISPR/Cas12a is a new RNA guided endonuclease that has recently been harnessed as an alternative genome editing tool. A team from the Department of Biological Engineering at MIT in Cambridge, Massachusetts, has pro- posed a high throughput multiplex nucleic acid detection microarray system. In their 2021 publication in Nature, the developers describe the system as potentially being the ultimate point of care diagnostic device, with high sensitivity and specificity, combined with high throughput, once integrated with upstream sample preparation and concentration stages. slow Rapid and highly sensitive approach for multiplex somatic fusion detection Somatic gene translocations are key to making an accurate dia- gnosis in many cancers, inclu- ding many sarcomas in children. Currently available molecular diag- nostic approaches to identify- ing somatic pathognomonic translocations ‘have limi- tations such as mini- mal multiplexing, high cost, complex compu- requirements, tational or turnaround times,’ said Samuel Abou, child cancer specialist at the Gustave Roussy Institute, Villejuif, France, and Brian Crompton, lab- oratory researcher specializing in the genomics of sarcomas at the Dana Farber Cancer Institute (Harvard Medical School, Boston, Massachusetts). This team has developed a new fusion detection assay optimized to mitigate these challenges. Their highly sensitive multiplexed digital PCR-based approach can identify the gene partners of multiple somatic fusion transcripts. This assay was valida- ted for specificity with cell lines and synthesized DNA fragments. Assay sensitivity was optimized using a tiered amplification approach for fusion detection from low input and/or degraded RNA. The assay was then tested for the poten- tial application of fusion detection from formalin-fixed paraffin-em- bedded (FFPE) tissue and liquid biopsy samples. According to results published in March 2022 in Modern Pathology, this multiplexed PCR approach was able to accurately identify the presence of seven different targeted fusion transcripts, with turnaround times of 1 to 2 days. The addition of a tiered amplifica- tion step allowed the detection of targeted fusions from as little as 1 pg of RNA input. ‘We have also demonstrated that the assay could easily be adapted for additional fusion targets,’ said Abou. This novel assay detects multiple soma- tic fusion partners in biological samples with low tumour content and low-quality RNA in less than two days. The inexpensive assay could be applied to surgical and liquid biopsies, particularly in plac- es with inadequate resources for  more expensive sequencing. Report: Bernard Banga Lab-on-a-chip The ongoing coronavirus pandemic caused by SARS-CoV-2 and its vari- ants is still a major public safety issue worldwide. Large-scale vaccination and sensitive detection are vital for preventing the spread of Covid-19 infection. To this end, teams from the Hospital of Guangzhou Medical University, and from the Department of Biomedical Engineering, Southern University of Science and Technology, Shenzhen, China, have established multiplexed lab-on-a-chip bioassays for testing antibodies against SARS- CoV-2 and its variants. ‘Virus neutralization assays that can measure neutralizing antibod- ies in serum are vital for deter- mining vaccine efficacy,’ said Prof Yong Xia, Department of Clinical Laboratory, Third Affiliated Hospital of Guangzhou Medical University. Compared with ELISA, their method demonstrates a low consumption of sample and reagents (10 μL), a low threshold for detection (0.08 ng/ mL), a rapid sample-to-answer time (about 70 min), and multiplexed ability (5 targets in each of 7 sam- ples in one assay). Results were published in 2022 in Analytical Chemistry. ‘We can also ramp up the throughput as needed,’ said Prof. Xingyu Jiang. His team performed a plaque reduction neu- tralization test (PRNT) for all vol- unteers. Compared with PRNT, their assay is fast, accurate, inexpensive, and multiplexed with multiple-sam- ple processing ability, which is good for large-scale serodiagnosis and vaccine evaluation. Sponsored · Safety blood collection set The gentle touch: Vacuette Evoprotect Evoprotect is the next stage in blood collection. The semi-auto- matic click mechanism protects the user from the risk of needle- stick injuries and makes the daily task of blood collection easier. Vacuette Evoprotect provides for gentle collection and safety from injury because we believe that every needlestick is one too many. The safety blood collection set meets current safety recommenda- tions and requirements and pro- vides substantial support for safe blood collection and infusion with the same set. User-friendly with an optimised design The Vacuette Evoprotect is ergo- nomically designed and intuitive to use. It has a winged needle, spe- cially developed for one-handed use. The potential needlestick inju- ries are minimized by the activa- tion of the safety mechanism while the needle is still in the vein. The Vacuette Evoprotect is suitable for a blood collection and a following infusion with the same set. triggers are located on the side of the product, facilitating one-hand- ed activation. The needle is then automatically retracted and, at the same time safely and irreversibly enclosed in the safety shield. A clearly audible click confirms that the safety mechanism is successful- ly activated. The safety mechanism is activated while still in the patient’s vein, offering the user an even high- er level of safety. The design of the safety mechanism promotes a straight removal of the needle from the vein, therefore offering more patient comfort. Maximum safety The Vacuette Evoprotect safety blood collection set has a semi- automatic safety mechanism. The Designed with patient comfort in mind The ultra sharp, triple bevel needle tip ensures optimal patient com- fort. A view window between the needle and the tubing gives an immediate visual indication that vein entry was successful. The tub- ing is designed to ensure that the needle stays as still as possible in the vein. The location of the safety mechanism triggers, on the side of the device, minimises the risk of the needle moving in the patient’s vein during activation. Versatile use The same device can be used for taking blood and for a subsequent infusion following afterwards. Using the same set for blood collection and infusion means a second puncture is not  needed. immediately More information: www.gbo.com or www.power4safety.com. EUROPEAN HOSPITAL Vol 31 Issue 4/22

6 Precise therapy PAT H O L O G Y / L A B O R AT O R Y Molecular Tumour Board to support better decision-making for patient care for targets As more genomic alterations become therapy, health institutions and hospitals are creating specialist Molecular Tumour Boards to support better decision-making for patient care. This evolving team, and its role, was highlighted in a presentation at the 34th European Congress of Pathology in Basel, Switzerland. Professor Fernando Schmitt, who is Head of the Molecular Pathology Unit (IPATMUP) in Porto, Portugal, told delegates that over the last decade several new substan- ces have been launched to treat solid tumours, with some receiving approval for multiple indications. With the advance in tumour molec- ular biology strictly associated with the need to include more molecu- lar analysis in tumour diagnosis, that has seen the concept of the Molecular Tumour Board (MTB) – first developed in Michigan in 2011 – play an increasingly pivotal role in the field, with patholo- gists as the pillars of these boards alongside clinicians. With 40 % of the oncology pipeline in Phase I trials, he forecasts major expansion of molecular targets over the next three to five years. Adequate sample Selection of an adequate sample for molecular analysis and under- standing the performance of the genomic tests is paramount, ena- bling integrations of morphology with molecular information, said Schmitt. He further underlined the importance of pathologists' exper- tise in the acquisition and man- agement of high-quality tumour samples. ‘This is important because tumour sampling represents a cru- SAFE AND SOUND Always effective. Never compromised. Aquasonic® 100 Ultrasound Transmission Gel Aquasonic CLEAR® Ultrasound Gel cial phase for any subsequent molecular analysis,’ he explained. Selection of the best samples has challenges and can be problematic where there are central laboratories, but he stressed the need to obtain good quality, and recent, samples and enable triage before sending them for more molecular analysis. Next-generation sequencing With samples of sufficient qua- lity, he said NGS (next-genera- tion sequencing) can be utilised to deliver precision medicine in oncology, adding that ‘pathology is essential for precision medicine.’ While ‘uniplex’ or ‘multiplex’ anal- ysis has a valid role, he indicated that NGS has advantages for analy- PARABEN FREE Visit us at MEDICA stand 9D41 AQUASONIC. DELIVERING THE PARKER PROMISE FOR QUALITY-ASSURED MANUFACTURING. PARKERLABS.COM MADE IN USA ISO 13485:2016 m o c . e b o d a . k c o t s – i o d u t s o t o f a t i u G © Molecular tumour boards should include pathologists, radiologists, molecular geneticists pharmacologists and clinicians. sis of a large number of genes, has good sensitivity, and is a modern tool that works well in pathology. Within the MTB arena, specific scales have been designed and approved for assessing samples for clinical actionability. And while a good quality sample ensures the material is suitable for molecular analysis, it may also deter onco- logists from migrating to liquid biopsy. ‘We know liquid biopsy is a good compliment for tissue anal- ysis but it is not a replacement for tissue analysis,’ said Schmitt. Multidisciplinary council The MTB expands the possibilities for genetics-guided therapy, but he emphasised the need to bring the required expertise together in the MTB ‘to ensure optimal care in an increasingly complex biomarker landscape.’ The aim of MTBs is to define the most appropriate strat- egies for genomic profiling and the correct interpretation of the results obtained from the molecular analyses of tumours. MTBs can and include specialitsts ranging from pathologists, radiologists, molecular geneticists to pharmacologists and clinicians. Reports should contain patient identification, and results that are concise and clear. Rare tumours The MTB allows integration of modern molecular diagnostic data and discussion of complex cases; it offers more precise and rational therapy recommendations; identi- fies and optimizes patients access to clinical trials; guides patient testing and treatment allocation practices; and promotes continuous medical education on emerging biomarkers. ‘It is a multidisciplinary deci- sion-making platform aimed to improve patient outcomes by con- sensus and standardized proce- dures,’ Schmitt added. ‘So, in the end we will have a treatment decision, we can request alternative molecular profiling assays, we can decide for an off- label treatment, or enrol a patient in a specific clinical trial.’ While he emphasized that all tumours should be discussed MTBs are particularly valuable for rare mutations, rare tumours without available therapeutic options, gene alterations that are not reported as associated with drug response or resistant, patients with onco- gene-addicted tumour, or tumours that do not show responsiveness to the available targeted therapies. Treatment costs Molecular guided therapy typical- ly takes up around 6 % of patient treatment costs, compared to hos- pitalization (34 %) and drugs (60 %). Schmitt also outlined the example of the POP-IPOP trial in Porto, looking at the challenges of broad molecular profiling such as timely access to results, access to the right drug at the right time, multiple potential actionable alterations in the same patient and of how to navigate the available evidence.  Report: Mark Nicholls Fernando Schmitt Fernando Schmitt is Professor of Pathology at the Medical Faculty of Porto University; he heads the Molecular Pathology Unit at IPATIMUP (Institute of Molecular Pathology and Immunology of the University of Porto), a Portuguese non-profit insti- tution dedicated to health sciences research. Furthermore, he is Director of RISE Health Research Network; and general secretary and president-elect of the International Academy of Cytology. U P C O M I N G E V E N T S I N PAT H O L O G Y A N D L A B O R AT O RY 3. -4. November 2022 16. International Conference on Advances in Clinical Chemistry and Laboratory Medicine, Amsterdam 7. - 8. December 2022 9th Digital Pathology & AI Congress: Europe, London 9. - 10. December 2022 16. International Conference on Laboratory Medicine and Pathology, London 6. - 9. February 2023 Medlab Middle East, Dubai 9. - 10. February 2023 Labquality Days, Helsinki EUROPEAN HOSPITAL Vol 31 Issue 4/22

8 Seven lessons R A D I O L O G Y Integration of AI in the clinical workflow In radiology, it is not about if but about when artificial intelligence (AI) will be used, said Professor Dr Tim Leiner of Utrecht University Medical Center at this year’s European Congress of Radiology in Vienna. For all those who are new to AI, the Dutch radiologist gave an over- view of the lessons he and his team have learnt so far. And indeed Professor Leiner offered seven distinct and entirely prac- tice-oriented lessons. Lesson 1: Identify the use cases ‘Identify the use cases. Artificial intelligence is not an end in itself; it is a means – the solution to specific problems. Focus on a pro- blem which you want to solve. Ask yourself where the use of AI really makes a difference.’ Lesson 2: Set up a team of experts ‘Set up a team of people with diverse expertise – among them obviously clinicians who contribute basic knowledge of the medical dis- cipline in question. However, you also need healthcare IT and infor- matics specialists, e.g. PACS admin- istrators. Moreover, I would recom- mend bringing AI experts on board since they can realistically assess the providers’ claims and promises. In addition, a patient representative is a good idea. Many algorithms are mere ‘black boxes’ and the exchange with a patient represent- ative can offer insights into patient psychology: some might feel uneasy about the technology's lack of trans- parecy and reject it outright.’ Lesson 3: Choose a platform ‘Choose a platform that fits into your environment. Take into account the long-term perspective: the range of available applications will grow. In Utrecht, we tried to be as open as possible. We wanted a platform where we decide what’s running on it, not the platform provider. It was crucial for us that the platform was able to process algorithms by dif- ferent providers as well as the algo- rithms we will develop in-house.’ Lesson 4: Choose a test product ‘After you have decided on a plat- form, choose a product that pro- mises to be a good solution for your specific clinical problems.’ Lesson 5: Define criteria of success ‘We recommend establishing spe- cific criteria beforehand that define when the implementation of an application is considered successful. Think about the goals you want to achieve with the application, e.g. higher throughput, higher preci- sions, saving time, or decreasing costs.’ Lesson 6: Conduct a pilot project in the hospital ‘Conduct a pilot project in your own institution. Never, ever buy an algorithm you did not test in your own clinical workflow. Don’t rely on demos. Collect as much feedback as possible from radiologists and clinicians. The test should run over a longer period of time, at least a few weeks, preferably some months. Assess the success of the pilot pro- ject using the criteria you defined I, Algorithm m o c . e b o d a . k c o t s – s o t i B y e g r e S © before the test. If you don’t do that you are setting yourself up for disappointment. Buy the algorithm only after a thorough assessment.’ Lesson 7: Implement smoothly ‘When you are satisfied with the platform and the product, imple- ment them smoothly. It is crucial to involve the users! Set up a team of ‘ambassadors’ who gathered expe- rience during the pilot project and can thus answer questions during and after roll-out.’ In addition to offering these lessons, Professor Leiner briefly addressed the obstacles that have to be sur- mounted when introducing AI in radiology. He pointed out that many algorithms were developed on the basis of very few cases: ‘Due to the General Data Protection Regulation and other privacy considerations the data sets are frequently small and biased. This is something we need to always keep in mind.’ Professor Leiner’s take-home mes- sage: ‘The introduction of artificial intelligence in healthcare is difficult  and hard work.’ Report: Michael Krassnitzer The 3 laws of robotics for AI in radiology Ethical considerations continue to fuel the discussion around artificial intelligence (AI). When science fiction author Isaac Asimov devised a set of rules for robotics in 1942, little did he know how relevant they would still be 80 years later. The rules, or laws, famously featured in the novel ‘I, Robot’, state that: • a robot may not injure a human being or, through inaction, allow a human being to come to harm; • a robot must obey orders given by human beings except where such orders would conflict with the First Law; and • a robot must protect its own existence as long as such pro- tection does not conflict with the First or Second Law. These three laws could also apply to AI, suggested Federica Zanca, Founder at Palindromo Consulting. ‘Radiologists should not trust an algorithm blindly and they should be aware of any issues related to the algorithm that have a negative impact on patient care,’ she said. For example, some AI tools might be subject to what is called exter- nal drifting of the model: the data input may change over time and therefore models developed on limited sample sizes may be sub- ject to degradation of the model performance over time. ‘Most AI models today are locked,’ Zanca said. ‘They have been trained with specific data to get a certain output and are not learning from new data to adapt accordingly.’ FDA regulation Solutions to remedy that situa- tion are in the pipeline and the FDA has already worked on re- gulation to approve self-learning tools. However, radiologists need to beware of self-learning models turning against them and errors sneaking in that compromise the wellbeing of patients, she insisted. ‘Quality assurance should be per- formed with each imaging medical device including AI software,’ sug- gested Zanca. ’ Gaspard d’Assignies, Co-founder and Director of the Medical Strategy at Incepto Medical, said that Radiologists need to know how to accept, reject or modify an algorithm. ‘We need to understand how these algorithms are going to modify patient care. So we need prospective, real-life studies that compare reader performance of man with machine vs. man and machine alone over time. should What happens when humans use AI? The central question that all these considerations revolve around, he added, is: How can AI improve workflow and accuracy according to the clinical context? ‘The design purpose must be very clear, and studies must be conduct- ed to validate these improvements.’ When creating an AI solution, a doctor should always be in the loop to check when the machine is right or not. ‘We don’t want a black box effect and we also must make sure that our algorithm is not biased and includes all the people it is supposed to represent,’ he said. One question, ten different values This is where the second of Asimov’s laws comes into play: AI should not obey the radiolo- gist when he or she is wrong. ‘If you asked a radiologist to do the measurement, you would possi- bly get ten different values. This may have consequences on treat- ment and the patient’s life. The machine is doing much better, and it always gives the same result for this type of measurements on the same image. For this indication, I think the machine is better suited,’ Zanca concluded. AI puts safeguards that may help prevent human errors, d’Assignies added. ‘For example, with a pul- monary nodule detection tool, the nodule threshold can't be set at 20mm, otherwise nothing will be detected,’ he explained. Radiation dose could be another instance where the machine should disobey an ill-advised order. ‘AI could warn when a radiologist would use too high a dose,’ Zanca said. ‘Today, we have a simple limit protection that is already embedded in the system. But AI could be used to drive exposure parameters.’ Recent reader performance studies focus on augmented AI, i.e., compar- ing human performance with the machine vs. human performance alone. However, radiologists should beware the risk of deskilling that comes with augmented AI; when professionals get used to a system doing something for them, that run the risk of losing the ability to perform this task themselves. Zanca ‘If the system fails, radiologists still need to be able to interpret images  without AI.’ Report: Mélisande Rouger More than just MRI accessories www.allmri.com Publisher Mediengruppe Oberfranken – Fachverlage GmbH & Co. KG E.-C.-Baumann-Str. 5 95326 Kulmbach/Germany Phone +49 9221 949-311 Fax +49 9221 949-377 Editor and Production Manager: Sonja Buske (SB) Editorial team: Wolfgang Behrends (WB) Sascha Keutel (SKE) Managing Directors: Walter Schweinsberg, Bernd Müller Founded by Heinz-Jürgen Witzke ISSN 0942-9085 Correspondents Austria: Michael Krassnitzer (MK) France: Jane MacDougall (JMD) Germany: Cornelia Wels-Maug (CWM) Dr Christina Czeschik (CC) Great Britain: Mark Nicholls (MN) Spain: Mélisande Rouger (MR) The Netherlands: Madeleine van de Wouw (MvW) USA: Cynthia E. Keen (CEK) Subscriptions Simone Sesselmann 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: Ralf Mateblowski Phone: +49 201 17755094 E-Mail: r.mateblowski@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 Maria Kaiser Phone: +1 250 726 4007 E-Mail: mkads@mac.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 31 Issue 4/22

1 0 R A D I O L O G Y First-incident ischemic stroke patients White matter hyperintensities in the brain: valuable biomarker to assess mortality risk White matter hyperintensities (WMH) on the brain seen on MRI represent a biomarker associated with a 50/50 risk of death within five years after a first incident acute ischemic stroke (AIS) or transient ischemic attack (TIA). Neurologists at the University of Bern have reported that prevalence and quantity of WMH are inde- pendently associated with mortali- ty, and that patients with moderate or severe WMH and additional chronic covert brain infarctions (CBI) had the highest risk of death in a study published in the Journal of Neuroimaging. The findings are based on an analysis of mortality outcomes over 8,179 patient years of 2,236 consecutive patients who received treatment for first incident AIS or TIA at the comprehensive stroke centre of Bern University Hospital. The study investigated both the association of different phenotypes (observable physical properties), the quantity, and the locations of WMH and CBI with death, and also to determine if the presence of both improved predictability. WMH – subcortical lesions that appear as increased bright areas in the brain – are known from prior research to be associated with increased risk of stroke, Alzheimer’s disease and other dementias, and death. Little is known about the phenotypes of CBI. researchers counted and The mapped the location of CBIs, which included large subcorti- cal infarcts, isolated grey matter lesions, combined grey and white matter lesions, and lacunes. They similarly counted and mapped the locations of WMH, as well as using the Age-Related White Matter Changes (ARWMC) scale to assess the severity of WMH. Using this classification scheme, they were able to differentiate between lesions that correspond to ischemic injury due to small-vessel disease and ischemic injury likely to be caused by embolism to the brain. All patients had undergone an MRI scan upon admittance to the hospi- tal or within one week of diagnosis. Their median age was 71 years (range 59 – 80), and 57 % were male. The patients were followed for a median of 4 years. The study did not track specific cause of death. Mortality rates Principal investigator Thomas Raphael Meinel, MD, of Bern University Hospital’s Department of Neurology, and colleagues repor- ted that patients without WMH had the least risk of dying, representing only 9 % of deaths. The mortality was highest for patients with mo- derate or severe WMH (ARWMC 2 or 3) and CBI, at 37 %, compared to 13 % for patients with minimal WMH (ARWMC 0 or 1) and no evidence of CBI. Mortality was not associated with CBI phenotype, location, or count. The higher the ARWMC score representing increased WMH, the greater the risk seemed to be of dying. Forty-nine percent of patients with ARWMC 3 rat- ings died, compared to 29 % with ARWMC 2, and 20 % with ARWMC 1. Twenty-nine percent of patients with any CBI died, compared to 17 % without any. ‘These brain frailty markers can be easily assessed without the need for advanced postprocessing,’ write the authors. ‘WMH can be reliably assessed on both routine 1.5 or 3 Tesla MRI and CT, and CBI on MRI. WMH are a more suitable surrogate marker than CBI for long-term mortality after stroke or TIA. Our study shows that 50 % of patients with severe WMH have a life expectancy of five years after the event.’ As deadly as cancer - but more pre- ventable ‘It is appropriate to communi- cate with high-risk cardiovascu- lar patients in a manner similar to communicating with oncology patients,’ Meinel tells European Hospital. ‘But unlike a cancer dia- gnosis, cardiovascular disease is highly preventable, and treatable by exercise, healthy diet, smoking cessation, blood pressure control, and medication. Significant risk for stroke recurrence ‘We tell patients with a high ARWMC score that there is a significant risk for stroke recurrence, and other major cardiovascular events inclu- ding death. It is also important to note that not all WMH are due to vascular risk factors, but possibly due to a genetic mutation, and that these people may be living a very heart-healthy lifestyle,’ he adds. ‘At Inselspital Universitätsspital Bern, we provide a detailed tai- lored package of interventions for each patient. We follow-up with the high-risk patients to ensure that the risk factors are well controlled and that the patient adheres to the  optimal medication.’ Report: Cynthia E Keen MORE CLARITY in cardiovascular imaging Due to an aging population, we are seeing a rising burden of cardiovascular diseases. Therefore we have developed sophisticated intraoperative imaging solutions that bring more clarity into cardiovascular interven- tions. Discover our innovative mobile C-arm solutions at www.ziehm.com/clarity. ALWAYS AHEAD m o c . e b o d a . k c o t s – 1 5 h a t u © Thomas Raphael Meinel Thomas Raphael Meinel, MD, is a an attending physician and group leader in the Stroke Center Bern of the Department of Neurology at Inselspital Universitätsspital Bern, Switzerland. His key research fields are stroke imaging, with an emphasis of covert brain infarction, futile inter- ventions in endovascular stroke treat- ment, anticoagulation and stroke, and stroke prevention. U P C O M I N G E V E N T S I N R A D I O L O G Y 3. -5.11.2022 Central German Neuroradiology Symposium, Dresden, Germany 19.11.2022 26. International Mammadiagnostics, online 27.11.2022 RSNA, Chicago, USA 12. -14.1.2023 IROS, Salzburg, Austria 1. -5.3.2023 ECR, Vienna, Austria 14. -17.3.2023 European Molecular Imaging Meeting, Salzburg, Austria EUROPEAN HOSPITAL Vol 31 Issue 4/22

1 2 R A D I O L O G Y Alternative to lumpectomy Breast cryoablation for surgically inoperative patients Breast cryoablation is an emerg- ing treatment for early-stage, local- ized breast cancer that destroys malignant tumours by freezing them. During the past decade, it has been increasingly utilized as an alternative to lumpectomy, but its long-term benefits compared to other breast cancer treatments are still unproven. Breast cryoablation is a non-inva- sive, less than 60-minute-long out- patient procedure performed with a patient awake under local anaes- thesia. The treatment causes fewer complications than breast-conser- vation surgery, has minimal impact on surrounding breast tissue, has less post-op pain management, and much better cosmesis. It is much less expensive, because it elimi- nates the need for operating room staff and equipment and requires less post-operative clinical fol- low-up. In addition to treating ear- ly-stage breast cancer treatments, it is used to destroy benign breast lesions and fibroadenomas, to treat patients with Stage IV metastatic breast cancer, and to treat patients unsuitable for surgery. Tips to make treatment easier Sarah E.H. Moorman, MD, a breast imaging fellow at the University of Michigan School of Medicine in Ann Arbor, presented the hospi- tal’s experiences with breast cancer patients for whom surgery would present a high risk. In addition to discussing patient demogra- phics and outcomes, she explained the procedure and offered tips to make the treatment easier for both radiologists and patients. Moorman reported that the findings of the study also showed that this was a safe and effective procedure for women who do not meet the crite- ria of the largest multi-institutional breast cryoablation clinical trial (ICE3) currently being conducted. The 16 patients in the study ranged in age from 57 to 94. All had comorbidities that made them poor surgical candidates, including prior stroke and serious cardiovascular conditions, advanced liver disease, COPD, emphysema, interstitial lung disease, and/or high risk of deep vein thrombosis and pulmonary embolism. To identify candidates for breast cryoablation, a surgical oncologist used surgical risk calcu- lators while conducting preopera- tive evaluations, and then discussed the cases with a radiologist. Discussing the odds In addition to surgical risks, the physicians discussed the odds of patients surviving longer with pro- active management of their breast cancer, and those for whom anti- endocrine therapy may not be effec- tive. None of the patients met the all-inclusion criteria used by the ICE3 trial, an ongoing clinical trial evaluating the safety and efficacy of liquid nitrogen-based cryoabla- tion in 194 patients aged 60+ years. The ICE3 participants had unifocal, ultrasound visible low-to-intermedi- ate stage invasive ductal carcinoma 1.5 cm or less in size, HR+/HER2-, and with no extensive ductal carci- noma in situ, invasive lobular car- cinoma, or metastatic disease. The Michigan patients had one or more tumour characteristics outside of the ICE3 criteria, including tumour size greater than 1.5 cm, invasive lobular carcinoma and/or DCIS, tumours less than 0.5 cm from the skin, and ER/PR negative tumours. LIKES SINGLE MODE, BUT IS ALSO PREPARED FOR MULTIPLE MODE. The Accutron® HP-D is the precise all-round talent for angiography and computer tomo- graphy. It is patient-friendly, reduce contrast agent, has optimised diagnosis, is intuitive to use and has dependable battery technology. Iceball from cryoablation freeze cycle (*) encompasses the malignant mass. Saline (stars) between skin & ice ball protects the overlying skin from injury from freezing temperature of cryoablation/underlying iceball Cryoprobe (arrowheads) positioned through the center of the malignant mass (*) Cryoablation is based on the cyto- toxic effects of cold that destroy cellular tissue, and consists of a first freeze, a passive thaw phase, and a second freeze. Cold temperatures freeze extracellular water, drawing water out of the cells and causing cellular dehydration. The cells swell and rupture during the passive thaw phase. The second freeze takes advantage of tissues that have been injured during the first freeze and conduct cold temperatures more efficiently, enhancing the damaging effects of cold and expanding the area of tumour necrosis. The procedure consists of using ultrasound imaging to guide a thin, needle-like device through the skin and into the breast tumour. The area is blasted with liquid nitro- gen for 6 to 8 minutes at -40 °C or lower, followed by a 10-minute thaw, and a second 6- to 8-min- ute freeze. ‘We identified several challenges with our patients,’ said Moorman. ‘Superficial tumours risk skin necrosis during cryoablation. Because saline can dissipate rapidly in pendulous fatty breast tissue, we used a 21G needle for rapid saline infusion, had a dedicated operator managing saline needle position- ing, and stocked the procedure room with extra bags of saline.’ She added: ‘Positioning of elderly patients can be difficult, especially for patients with large breasts, neck mobility issues, and the inability to extend an arm above their head. We recommend allowing extra time for the procedure, using a wedge for patient positioning and com- fort, keeping the breast in position with paper tape, and placing the patient’s arm at their side or across the abdomen. Patients with comor- bidities such as COPD or congestive heart failure may have shortness of breath and decreased oxygen, and risk hypoxia and fluid over- load. We provide oxygen to these patients and elevate the head of the bed by as much as 45°. Targeting of the mass is likely feasible with needle entry from lateral, medial, or peripheral breast, but requires about 5 - 6 cm of tissue depth from surface,’ she explained. Moorman advised that fat necrosis is expected to be seen on follow-up imaging, and that any changes near the cryoablation cavity suspicion findings should be promptly biop- sied. All of the patients recovered rapidly, and none experienced seri-  ous side effects. Report: Cynthia E. Keen EUROPEAN HOSPITAL Vol 31 Issue 4/22

1 4 H E A LT H I T Sponsored · A tool to empower healthcare organizations Eliminating silos and improving patient outcomes improving consistency in workflow management. Users of Ovid Synthesis are fin- ding the solution to be a powerful resource in terms of collaboration, standardization, and promoting alignment across different depart- ments within their organizations as well as in other geographical areas. They also find it to be a transfor- mative tool by allowing new clini- cians to adapt to evidence-based practices within their hospitals. Still, without executive leadership emphasizing the core tenets of a culture of quality improvement, no tool can deliver meaningful results by itself. Communication, observation, and education are the main drivers of change manage- ment, and silos will remain if those values are not reflected by hospital leadership. Health system CEOs and executive teams will be the champions of change and must encourage a culture of transforma- tion to drive new enhancements. With support from the top, sus- tainable change can be achieved by starting with an evidence-based approach and integrating tactics to engage hospital staff, increase employee morale and confidence in the organization, and secure buy- in from prominent stakeholders. Hospitals can make a significant impact on quality improvement and performance by combining an evidence-based methodology with proven purpose-built technol-  ogies. Report: Vikram Savkar Vikram Savkar Vikram Savkar is the Senior Vice President and General Manager, Medicine Segment of the Health Learning, Research & Practice busi- ness at Wolters Kluwer. Previously, Mr. Savkar served as General Manager for several businesses in the Legal & Regulatory division before join- ing the Health Division in late 2019. Prior to joining the company, he held senior positions at Nature Publishing Group and Pearson Education and has earned degrees in Physics and Classics from Harvard University. from the Quality Improvement Conference in Gothenburg, Sweden, held in June of 2022 showed that the largest concerns that organizations presently have are a lack of stan- dardization and workflow silos. The unnecessary duplication of efforts and unsupported decision-making stemming from these issues can have a significantly negative impact on patient care and outcomes, as well as on the financial health of the organization. Healthcare insti- tutions are finding that they need to implement innovative technolo- gies that break through deep-seated silos, establish successful strategies, and enable rapid decision making. To facilitate change management and a progression of continuous quality improvement, a workflow tool that helps makes processes more efficient, successful, and col- laborative is needed. One example is Ovid Synthesis Clinical Evidence Manager, developed at Wolters Kluwer, a tool that empowers healthcare organizations to foster an innovative culture by reduc- ing duplicative efforts, promoting collaboration, and streamlining research. The cloud-based work- flow management solution offers a centralized dashboard view of all quality improvement research projects within an organization, Clinician and nurse burnout is a frequently discussed topic in the healthcare community in the wake of the Covid-19 pande- mic. According to an analysis of the NHS published earlier this year, more than 400 workers in England have left the NHS to restore work-life balance within this past year. Burnout and cases of post-traumatic stress disor- der after working through the Covid-19 pandemic are major causes of the departures. Less widely discussed is the impact staff burnout and other conse- quences of the Covid-19 pande- mic are having on the quality of care that hospitals deliver. Healthcare institutions are con- stantly aiming to ensure high-qua- lity patient outcomes, attain maxi- mum efficiency, and reduce health- care-related costs. This is always a challenge; in this age of staff- ing disruptions, it is even more so. Finding systematic solutions to improve healthcare quality, there- fore, is a key focus for healthcare systems today. Hospitals and health systems are finding that they can no longer approach quality improvement (QI) in a dispersed manner. Survey results Sponsored · Clinio series All-in-one computer for use in hygienically sensitive areas Nowadays, the operating thea- tre is networked and staff have access to a wide range of infor- mation and image data. With the Clinio, Rein Medical offers a medical all-in-one (AIO) comput- er that has been designed for use in hygienically sensitive areas and can be used both as a thin client and as a full PC system. ‘With the Clinio, users can retrieve and process data from medical information and imaging systems such as HIS, RIS, PACS, PDMS or operating theatre image manage- ment. We only use highly available components that are well-known in the industry. These are particu- larly powerful, temperature resis- tant and consume little power,’ says Gernot Schnock, Development & Product Management at Rein Medical. This makes the Clinio extremely extremely low-maintenance and designed for continuous 24/7 use. With its VESA 100 mounting device, the AIO computer can be individually wall mounted, attached to cei- ling lamps or installed on mobile and stationary workstations. robust, Hygiene is a top priority As the Clinio is designed for hygienically sensitive areas, it can be used not only in the operating theatre but also in the central ste- rile supply department, the emer- gency room, intensive care and IMC wards, the shock room, during mobile ward rounds on a trolley or as a documentation workstation. The enclosure is made entirely of aluminium; it is hermetically sealed and fanless to prevent active air exchange with the environment as well as swirling of dust and germs in the surroundings. ‘The IP65- protected front and the surface coated with the germicidal powder coating Polyflex Steridur II allow for easy wipe disinfection. The Clinio is resistant to listed surface disinfectants of the VAH disinfec- tant list,’ Schnock explains. The optimised housing shape has smooth surfaces and no cooling fins, ensuring safe and hygie- nic preparation. The glass pane is fitted flush and sealed as well as circumferentially protected against mechanical damage. The protective glass made of micro- etched glass features a high anti- reflective effect and non-sparkling effect (anti-glare). The hygienic flap of the connection panel with IP54-certified cable feed-through can be opened without tools. The Clinio is operated via the com- mand bar, a hygienic and wear-free control panel behind glass that can also be operated with surgical gloves. ‘This is used, for example, to switch the computer to stand-by mode and to activate or deactivate the touch or keylock. The device automatically goes into keylock status after start-up, so the touch on/off function remains usable for cleaning purposes. If keylock is switched off manually, the device remains in this status and enables permanent operation via the keys,’ explains Schnock. If they want to, users can also enter data using a stylus. Even more optional possibilities To make data collection easier, a barcode reader with omni-direc- tional scanning function can be integrated on the underside of the Clinio, as well as a multi-frequency RFID reader behind the protective glass. The optionally mountable soundbar with DSP sound proces- sor delivers a clear reproduction of audio files and is optimised for speech reproduction. The devices of the Clinio AIO series are scal- able in terms of screen size from 21.5" to 27", performance class, storage capacity and much more.  • Class 1 medical device according to MDR (EU) 2017/745 with strict conform- ity assessment procedure. • Certified according to EN 60601-1-2:2015 • Certified according to EN 60601-1:2013-12 • Emission/interference resistance according to EN 55032:2015 Class B / CISPR32:2015 • Risk management report according to EN ISO 14971 • Usability test EN 60601-1-6 • EUDAMED registration With this, Rein Medical offers operators and system integra- tors easy integration and a spe- cial level of safety when cre- ating medical systems in the near-patient environment. EUROPEAN HOSPITAL Vol 31 Issue 4/22

1 6 S U R G E R Y Assistance and decision-making systems Artificial intelligence for preoperative planning in surgery In surgery, artificial intelligence (AI) is applied mostly in ima- ging, navigation, and robotic intervention. However, AI can also play a major role in preop- erative planning. Objective deci- sions-making, optimal utilisation of operating theatres and less overtime are additional advan- tages that are achieved with the use of AI in surgery. Surgeons make complex, high-risk decisions with significant impact on patients’ lives, usually under time pressure. The multitude and variability of diagnostic data and the potential dangers of a surgical intervention pose tough challenges for the staff. Therefore, thorough pre-operative planning, in which surgeons prepare the operation based on medical records, histori- cal methods and data, is essential for the success of a surgery. influence Decision-making systems Surgeons decide based on their training, experience, and know- ledge, so their judgement or bias can them. Decision- making systems based on AI have the potential to overcome these limitations. These systems enable the processing of many patient data sets, which can be transferred into mathematical models and enriched with background knowledge to recommend or advise against a certain therapy. Artificial intelligence is used as a decision-support tool with two tasks: to compensate for the diffe- rent levels of knowledge and expe- rience of surgeons by objectively and individually analysing patient data; and highlighting clinically Sonos 12 Ultrasound system relevant correlations that the phy- sicians may not recognise. example, scientists For at Heidelberg University Hospital are developing a ‘Cognitive Medical Assistant’ (KoMed). The aim of the project is to recognise patterns from laboratory results, image data, diagnoses, and therapies and to identify correlations that can create individual risk profiles. The algo- rithm should identify a patient's surgical risk before an operation, facilitate therapy decisions and prevent complications. Specifically, the algorithm is designed to pre- dict the likelihood that patients will need a blood transfusion. In another study, researchers used deep learning to predict kidney failure, mortality and post-opera- tive bleeding after heart surgery. With their approach, the scientists achieved superior results com- pared to standard clinical refer- ence tools. Limitations Clinical trials in AI-assisted sur- gery with evidence are scarce, and AI-assisted decision support has yet to realise its potential. Currently, decision support appli- cations are limited to risk stratifi- cation of surgical treatment or its complications. In surgical onco- logy, AI studies are mainly limi- ted to the feasibility of predicting oncological outcomes, postopera- tive complications and predicting mortality and survival. For now, only a few comparative studies focus on improving treat- ment decision-making. Commercial attempts to address this pro- Features: 19” HD LED monitor Rotatable: ± 90° for optimal display of Images 10.1” HD customizable touch panel with multi angel visualization Innovative energy-saving light for ergonomic workplace Single crystal technology as the latest technology in transducer Wide angle TV probe up to 210° Volumetric real time 4D Left ventricular tracking Auto breast lesion detection Smart HIP for pediatric hip scanning Quantitative Elastography Integrated Gel Warmer Smart Standby Mode for instant power on mode Internal Battery for uninterrupted Service even in absence of AC power www.sternmed.de email@sternmed.de +49 751 35 97 8-0 m o c . e b o d a . k c o t s – t o h s e m a r f y h w © AI in surgical planning can improve the treatment pathway. blem have struggled with various issues. IBM Watson for Oncology Cognitive Computing System drew on data, test cases and experts at Memorial Sloan Kettering Cancer Center to generate treatment re- commendations. However, studies have so far shown little consisten- cy between Watson’s treatment re- commendations and actual surgical treatments. Google’s DeepMind, on the other hand, was criticised for its lack of transparency and pro- blems with data protection. OR planning Scheduling operations, assigning patients, surgeons, nursing staff and anaesthetists – surgery plan- ning in hospitals is a dynamic, complex, and time-consuming undertaking. Emergencies, staff absences or unforeseen events du- ring a procedure are common short-term events that can quickly upset the schedule. Even for expe- rienced OR managers, exact plan- ning or possible manual adjust- ments are therefore difficult. The results are not only suboptimal uti- lisation of operating theatres and thus of resources but also inflict additional strain on staff. Assistance management systems that schedule staff, treatments, times, resources as well as upstream and downstream processes around the operating theatre with the assistance of AI could provide a remedy. Based on the collection and analysis of data, these systems create a scheduling plan according to definable criteria, evaluate what is happening and adjust the plan in real time if unexpected events occur. A stable OR plan not only relieves staff in the long term by elimi- nating shifts and overtime, but also increases treatment quality because the staff has more time to care for the patients. Optimised capacity also pays off for a hos- pital's economic situation by ena- bling better scheduling of staff and bed occupancy. In the ‘KIOM’ project, researchers are developing an interactive AI-supported assis- tance system for efficient planning of operations. It is intended to interactively support OR managers in scheduling surgeries. The com- Further areas of application involve AI-based estimation of the correct opioid dosage and the identifica- tion of patients who could benefit from pre-operative counselling by a hospital's acute pain service. t c e j o r p A R D N A S S A C © AI-assisted anaesthesiology. bination of mathematical optimisa- tion approaches with AI methods generates stable and transparent OR plans. Anaesthesiology Many studies test AI applications for anaesthesiology in clinical set- tings. The aim is to predict poten- tial complications with a personal- ised perioperative therapy and to sustainably improve the postope- rative outcome of patients. In the Cassandra project, research- ers are developing and evalua- ting a virtual ‘AI team member’ to support anaesthesiologists in the treatment and care of patients. Also, they aim to reduce the time and mental burden on staff and improve documentation quality through automatic documentation. In addition, researchers are explo- ring the potential for AI to con- trol the mechanical ventilation of anaesthetised patients. Other trials study how the AI can automate weaning from ventilation. Many AI software developments and studies on their applications show improvements over existing methods. But the hype around AI can lead to unrealistic expec- tations and disappoint clinicians and patients. Assistance and deci- sion-making systems can give hos- pitals greater planning certainty and improve surgical outcomes, but operators should use them in the right situation to solve a speci- fic problem. The examples of IBM and Alphabet show how complex the integration of AI systems into the clinical workflow can be.  Report: Sascha Keutel EUROPEAN HOSPITAL Vol 31 Issue 4/22