AI models detect genetic changes in colorectal cancer from tissue images

For digital health at the TUD Dresden University of Technology, an international interdisciplinary research team led by Professor Jakob N. Kather of the Else Kröner Fresenius Center (EKFZ) analyzed seven independent cohorts of patients from Europe and the United States using a newly developed AI model. This model detects genetic changes and resulting tissue changes in colorectal cancer directly from tissue cross-sectional images. This will allow for faster and more cost-effective diagnosis in the future. Data and computer science, epidemiology, pathology, and oncology experts worked closely to develop, validate and analyze the models. This study has been published in the Journal of The Lancet Digital Health.

The multicenter study analyzed approximately 2,000 digitized tissue slides from colon cancer patients across seven independent cohorts in Europe and the United States. The samples included full-length images of tissue samples as well as both clinical, demographic and lifestyle data. Researchers have developed a new “multi-target transformer model” to predict a wide range of genetic changes directly from periodically stained tissue cancer tissue sections. Previous studies were usually limited to prediction of single genetic changes and did not explain co-occurrence mutations or shared morphological patterns.

Previous deep learning models and analysis of underlying tissue changes generally focus on only a single mutation at a time. However, new models can simultaneously identify many biomarkers, including those not considered clinically relevant. This could be demonstrated in several independent cohorts. We also observed that many mutations occur more frequently in microsatellite invasive tumors (MSIs).

Marco Gustav, M.Sc., First author of EKFZ research and researchers for digital health in Tu Dresden

Certain types of colorectal cancer can be classified based on microsatellite instability (MSI). Microsatellites are short, repetitive DNA sequences that span throughout the genome. In cancer, MSI can occur when these sequences become unstable due to defects in the DNA repair system. MSI is an important biomarker for identifying patients who may benefit from immunotherapy. “This suggests that different mutations contribute collectively to changes in tissue morphology. The model recognizes shared visual patterns rather than independently identifying individual genetic changes,” he adds.

The researchers demonstrated that their models partially surpassed the established single target model in predicting numerous biomarkers, such as BRAF, RNF43 mutations, and microsatellite instability (MSI) directly from pathological slides. The pathological expertise required to assess tissue changes from histological slides was provided by experienced medical professionals. Dr. Nic Reitsam of Augsburg, University Hospital, played an important role in this study.

Emphasizing the importance of the research, Jakob N. Kather, professor of clinical artificial intelligence at EKFZ for digital health and senior oncologist at NCT/UCC, Carl Gustav Carus Dresden, a senior oncologist at NCT/UCC, stressing the importance of the research, said: cancer. In the future, this technique can be used as an effective prescreening tool to help clinicians select patients for further molecular testing and guide their personalized treatment decisions. ”

The researchers are currently planning to extend this approach to other types of cancer.

This study was conducted through interdisciplinary collaboration among numerous scientists from major research institutes in Europe and the United States. In addition to TUD and Dresden University Hospital, partners included the Medical School at the University of Augsburg, the National Centre for Tumor Diseases (NCT), the Fred Hutchinson Cancer Centre in Seattle (USA), the Medical University of Vienna (Austria), and the Mayo Clinic (USA).