Fraunhofer Institute for Production Technology IPTResearch project "KI4Med"
The evaluation of tumors during surgery is complex, expensive, and time-consuming. The conventional method, in which tissue samples are taken during surgery and immediately examined manually by specialists, leads to longer waiting times and higher costs. This costs patients valuable time, especially in the case of cancers such as colon, liver, breast, or prostate cancer.
The examination of a removed sample takes about 20 to 40 minutes. These long waiting times during surgery cause additional stress under anesthesia. The Center for Cancer Registry Data at the Robert Koch Institute estimated around 500,000 new cancer patients in 2020. Assuming a similar number each year, these waiting times could cost German hospitals an estimated 200 million euros per year. The Fraunhofer IPT's goal is therefore to develop a faster method that significantly reduces the time required for tissue examination during surgery, relieves the burden on staff, and is more patient-friendly.
Artificial intelligence in the operating room
In the project "Artificial Intelligence in Medical Imaging (KI4Med)", Fraunhofer IPT and Fraunhofer Austria are jointly developing a method based on optical coherence tomography (OCT) that reduces the time required for tissue sampling to a few seconds by automating image analysis. The aim is to detect tumors faster, more objectively, and more precisely than before. Specialized personnel no longer need to be on site for the evaluation, and the associated subjectivity in the assessment of results is eliminated.
Innovation based on proven foundations
The collaboration between Fraunhofer IPT in Germany and Fraunhofer Austria Research GmbH in Austria builds on earlier work in which OCT results were successfully interpreted automatically. Optical coherence tomography offers great potential for non-invasive imaging in medicine and for non-destructive testing.
As a result of the project, the two Fraunhofer partners are developing algorithms that can accurately distinguish between healthy and diseased tissue based on OCT data. These algorithms are designed to be highly sensitive and specific to detect various clinical pictures in the future.
Generation of synthetic OCT data
Another goal will be algorithms that generate synthetic OCT data based on existing data to expand the amount of training data for AI. This will integrate OCT data synthesis for complex layer structures and models for the first time. In this way, it will be possible to simulate different tissue types and thus obtain realistic data sets.
The synthesis of OCT images creates a virtually infinite database. On this basis, data-driven approaches to artificial intelligence and big data can be pursued even when there is a shortage of real samples. Exemplary tissue samples are indispensable when the number of cases is limited and the associated pool of images is therefore limited, as is common in medicine.
Project partners
- Fraunhofer Austria Research GmbH, Vienna (Austria)
- Fraunhofer Institute for Production Technology IPT, Aachen