Fraunhofer Institute for Digital Medicine MEVISInstitute News
You are cordially invited to visit us at RSNA 2023. You will find us in Booth 2008, South Hall, Level 3 at McCormick Place, Chicago.
At this year’s annual RSNA meeting (Radiological Society of North America) in Chicago, Fraunhofer MEVIS presents, among others, the following promising innovations for AI-supported medicine: SATORI which helps to conduct multicenter studies and OncoChange which supports the follow-up of tumor treatments.
Whether for early detection, diagnosis, or therapy – artificial intelligence becomes increasingly important in medicine. One example for its application in radiology is segmentation. This process involves the identification of different organs or organ parts in medical image data, such as a CT scan. An AI system can complete this time-consuming task within seconds and thus, significantly reduce the workload on staff.
Ideally, such algorithms are developed in multicenter studies. “Especially in the case of rare pathologies, multicenter studies are necessary to acquire a broad database to train the algorithms,” emphasizes Fraunhofer MEVIS researcher Bianca Lassen-Schmidt.
However, conducting such studies is complex: different clinics work with different file formats and computer programs. This is where the SATORI software toolkit comes in. It processes image data from different clinics, makes them comparable and segments the anatomical structures and pathologies in the images. The training of a new AI system can then begin directly from SATORI.
SATORI has been successfully deployed within RACOON – an association of radiology departments of all 38 university hospitals in Germany, which originated during the coronavirus pandemic and has continued its efforts since. Its concept: each site hosts a server with identical software – including the Fraunhofer MEVIS program SATORI. The image data is analyzed locally at first. Next, the results from the individual clinics are being aggregated and evaluated on the RACOON central server. In the first phase, experts from the network analyzed pathological changes in the lungs, followed by other data such as heart MRIs.
The project is now being expanded to numerous other medical issues – and the Fraunhofer MEVIS software SATORI is ready for this: “SATORI is highly configurable,” says Lassen-Schmidt. “We can adapt the system to specific questions, allowing us to process exactly what a research team has in mind.” At the RSNA, Fraunhofer MEVIS will be showcasing the possible applications of SATORI in the RACOON network using a demonstrator.
The second system, OncoChange, presented by the Fraunhofer MEVIS experts, addresses tumor progression monitoring during cancer therapies. In practice to date, medical staff are faced with several challenges. For example, the results of the tumor measurement often depend on the personnel involved. In addition, it is not always easy to localize a tumor on follow-up images.
OncoChange is a complete solution supporting both tasks. It can precisely measure the tumors and if the patient gets examined again after the initial treatment, OncoChange can find the tumors again, determine their size and allow a more reliable assessment whether the therapy was successful or not. The system thus relieves the burden on doctors and makes their work more efficient. “As a result, radiologists have to spend less time dealing with the images and have more time for their patients, for example to discuss next treatment steps,” says Fraunhofer MEVIS researcher Bernd Kümmerlen.
At the RSNA, the Fraunhofer MEVIS team is presenting a demonstrator that unleashes the potential of AI: individual components of OncoChange, such as the precise tumor segmentation or the progression registration, can be integrated into existing systems like PACS viewers. Fraunhofer MEVIS is currently negotiating with several industry partners to integrate individual OncoChange components into their products.
Other topics that Fraunhofer MEVIS will be presenting at the RSNA:
- gammaSTAR: Vendor-independent platform for MR imaging
- Navigation and tracking for endovascular interventions
- Algorithms for segmentation and registration
- mAIvalve: Virtual intervention planning to improve mitral valve therapy
- HistokatFusion: Histopathologic registration