For healthcare professionals. For you. 10 MED Komponenten Konstante Drehzahlen in Arzneimittel-Testgeräten 17 MED Software Digitale Messung des Blasenfüllstands 30 MED Elektronik Sounddesign für medizinische Geräte 3-4 | 2024 07 Titelstory Qualität, Sicherheit und Ressourcenschonung vereinen € 22.- The Guide to Imaging Technology and Informatics in Europe RADBOOK 2024 Vol. 17 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 4 2 0 2 r e b m e v o N November 2024 93. Jahrgang ku-gesundheitsmanagement.de 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- of the Institute of Medical In- formatics, University of Lübeck, formatics, University of Lübeck, and Head of AIMedI explains: and Head of AIMedI explains: “Our aim is to support clinicians “Our aim is to support clinicians with the classification of cancer with the classification of cancer tissue of malignant brain tu- tissue of malignant brain tu- mours such as glioblastoma. This mours such as glioblastoma. This also helps with the volumetric also helps with the volumetric measurement of a brain tumour measurement of a brain tumour for which we need to be able to for which we need to be able to define which pixels form an ob- define which pixels form an ob- ject.” ject.” www.healthcare-in-europe.com Grün: Peritumorales Edem; Gelb: Gadolinium-verstärkter Tumor 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 tonomously learn from training data. To facilitate this, the al- data. To facilitate this, the al- gorithm requires original images gorithm requires original images that must be manually marked that must be manually marked by an expert to create a so-called by an expert to create a so-called annotated image. annotated image. This approach has already been This approach has already been successfully used in many areas successfully used in many areas of public life, such as for the of public life, such as for the analysis of traffic, where non- analysis of traffic, where non- specialists perform the an- specialists perform the an- 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- dels. However, deep learning al- gorithms deliver more reliable gorithms deliver more reliable results, the more data is avail- results, the more data is avail- able. able. Using U-Nets for optimised results To overcome the constraints of To overcome the constraints of limited manually annotated train- limited manually annotated train- ing data, scientists in the AIMedI ing data, scientists in the AIMedI group have trained a U-Net, a group have trained a U-Net, a convolutional neural network convolutional neural network developed seg- developed seg- image image for for k c e b ü L f o y t i s r e v i n U © 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 At the heart of this endeavour is the aspiration to generate an- the aspiration to generate an- notated synthetic brain tumor notated synthetic brain tumor images to free up experts’ time, images to free up experts’ time, which they anyhow only unwill- which they anyhow only unwill- ingly bestow on annotating 3D ingly bestow on annotating 3D datasets. Handels explains: “We datasets. Handels explains: “We aim to generate new data, so- aim to generate new data, so- called synthetic image data, called synthetic image data, where the labeled data is auto- where the labeled data is auto- Heinz Handels n i z a g a m h c a F s a D – t n e m e g a n a m s t i e h d n matically generated, using gener- u ated net- s e works (GANs) or diffusion G U models.” However, this type of K network poses its own chal- lenges like its substantial storage requirements. Prof Dr Heinz Handels Director of the Institute of Medical Informatics, University of Lübeck adversarial 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. ■ Article: Cornelia Wels-Maug CONTENTS RADIOLOGY LABORATORY AI RESEARCH 2 5 7 11 G K . www.healthcare-in-europe.com o C & H b m G e g a l r e v h c a f o g m MEDIZIN- UND MEDIZIN- UND KRANKENHAUSTECHNIK KRANKENHAUSTECHNIK Auf dem neusten Stand EMPLOYER BRANDING EMPLOYER BRANDING EMPLOYER BRANDING Wertvolles Gut Wertvolles Gut Wertvolles Gut € 22.– Your guide to laboratory and pathology equipment in Europe 2024/2025 Vol. 11 Automation & Sample Processing Chemistry & Immunochemistry Hematology Pathology DNA Microbiology POCT Information Technology Other Applications NAEOTOM Alpha®, the world’s first photon-counting CT, is nothing less than the reinvention of computed tomography. NAEOTOM Alpha®, the world’s first photon-counting CT, is nothing less than the reinvention of computed tomography. Benefit from a range of clinical options and breakthrough consistency never seen before in conventional CT scanners, Benefit from a range of clinical options and breakthrough consistency never seen before in conventional CT scanners, based on the revolutionary direct signal conversion of photon-counting CT. NAEOTOM Alpha. CT redefined. based on the revolutionary direct signal conversion of photon-counting CT. NAEOTOM Alpha. CT redefined. Discover your topics in our healthcare journals. We offer a broad range of publications in the field of healthcare. mgo-fachverlage.de/healthcare mgo fachverlage GmbH & Co. KG E.-C.-Baumann-Str. 5 · 95326 Kulmbach · Germany