Molecular tumor classification through AI

For targeted colorectal cancer therapy, the treating physicians need information about the tumor’s molecular subtype. Computers could analyze digital tissue images for this purpose.

A more efficient method for colorectal cancer prognosis

For targeted colorectal cancer therapy, the treating physicians need information about the tumor’s molecular subtype. Computers could analyze digital tissue images for this purpose. The method was developed by a research team from Zurich University Hospital and Oxford University.

Colon cancer is the third most common tumor disease in men and women with approximately 1.8 million new cases worldwide every year. With precise information about the molecular subtype of the tumor using RNA sequencing, personalized therapy can be supported. Patients with particularly aggressive tumors can be better identified and molecularly classified. However, this is resource-intensive and expensive. In addition, up to 20% of samples cannot currently be conclusively classified because, for example, too little material is available or the results are ambiguous.

Further development thanks to image analysis and artificial intelligence

A research team led by Prof. Dr. Viktor Kölzer, Institute of Pathology and Molecular Pathology at Zurich University Hospital, and Prof. Dr. Jens Rittscher, Institute of Biomedical Engineering at Oxford University, has now developed a substantially cheaper and faster method: they have computers analyze high-resolution images of histological sections with artificial intelligence. In this way, they learn about the gene expression profile of the tumor and are given indications as to which drug can be used to treat it.

In contrast to the previous gold standard - RNA sequencing - no further tissue material is required for this purely image-based procedure. It also works on very small tissue fragments and allows the classification of previously inaccessible tissue samples due to the technical limitations of sequencing. The process also potentially generates significantly lower costs. Image-guided methods have the potential to revolutionize the personalized therapy of colorectal cancer. However, the new technology requires the appropriate preparation of histological sections: "In order to be able to use artificial intelligence for tumor analysis, we have to digitize the pathology," said Dr. Kölzer.

Strategically important for personalized medicine

In April of this year, Dr. Kölzer took up the professorship for computer-aided image analysis in pathology at the USZ. According to the researchers, the first chair of this kind in Switzerland is of great strategic importance for personalized medicine. Dr. Kölzer began his work on the AI-assisted procedure during a stay at Oxford University, where he found broad interdisciplinary support from pathologists, bioinformaticians, clinicians and statisticians from various institutes and centers in the multi-institutional Stratification in Colorectal Cancer Consortium. 

For the study, 1,553 scans of tissue sections were analyzed using the latest methods of machine vision and artificial intelligence with RNA expression profiles, gene mutations, and clinical progression data. The new method must now be validated in prospective, randomized clinical trials. Pathologists are still working largely analogically - but this could change, even in countries with fewer resources. Kölzer said: "After validation, it would be possible to centralize the classification of colorectal tumors and make the technology available". Scans of histological sections could be sent to university centers, where they could be evaluated and the results communicated electronically. In the long term, the method could also be used for other tumor types and even for other diseases.