An international research team led by Prof. Jakob N. Kather from the Else Kröner Fresenius Center (EKFZ) for Digital Health at TU Dresden has developed a novel AI model that could speed up the diagnosis of colorectal cancer and make it more cost-effective. The model detects genetic changes and resulting tissue changes directly from tissue sectional images, as shown by a multicenter study with almost 2,000 patient samples from Europe and the USA. The results were published in The Lancet Digital Health .

The study analyzed digitized tissue sections of colorectal cancer patients from seven independent cohorts, supplemented by clinical, demographic and lifestyle data. The new “Multi-Target Transformer Model” simultaneously predicts multiple genetic changes, such as BRAF or RNF43 mutations and microsatellite instability (MSI), from standard stained histological sections. In contrast to previous models, which usually detected only one mutation, this model also identifies previously clinically ignored biomarkers and recognizes common visual patterns that arise from the interaction of several mutations. In particular, MSI, an important biomarker for the suitability of immunotherapies, could be reliably detected.

The development of the model was based on an interdisciplinary collaboration of experts from data science, computer science, epidemiology, pathology and oncology. Pathological expertise, for example by Dr. Nic Reitsam from the University Hospital Augsburg, was decisive for the assessment of the tissue changes. The model performed at least as well as established approaches in predicting biomarkers and in some cases exceeded them. It showed that many mutations occur more frequently in MSI tumors, which provides new insights into the relationships between molecular and morphological changes.

In the future, the technology could serve as a pre-screening tool to specifically select patients for further molecular tests and to optimize therapy decisions. The researchers plan to transfer the approach to other types of cancer. Renowned institutions such as the National Center for Tumor Diseases in Heidelberg, the Fred Hutchinson Cancer Center in Seattle, the Medical University of Vienna and the Mayo Clinic were involved.

The EKFZ for Digital Health, funded by the Else Kröner-Fresenius Foundation, is driving such innovations in order to use digitization in medicine for better health care. The study marks an important step towards faster, more precise and cost-efficient cancer diagnostics.

Original Paper:

Assessing genotype−phenotype correlations in colorectal cancer with deep learning: a multicentre cohort study – The Lancet Digital Health

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability. ected.