Artificial Intelligence-based analysis of histopathology images of biopsies to predict response to neoadjuvant therapy

Project summary

Erasmus MC and Ellogon.ai will collaborate to use artificial intelligence in microscopic images to select bowel cancer patients who benefit from chemotherapy.

In the Netherlands, approx. 13,500 patients are diagnosed with colorectal cancer annually, while approx. 4,500 people die of (colo)rectal cancer every year, and approx. 10,500 patients are treated surgically for this disease (latest NKR data (2021)). This surgical treatment is associated with considerable side effects, while in up to 25% of patients, the use of chemotherapy prior to surgery can result in a complete disappearance of the tumor. This project will provide a novel tool to better identify patients in which surgery should be omitted, thereby avoiding major surgery. Therefore, this project will increase the quality of life of these patients, and will also save valuable resources by avoiding unnecessary chemotherapy and surgery.

The proposed state-of-the-art technology is able to provide AI-assisted prediction of chemotherapy response for a variety of common cancers, solely based on the morphology in easily and cheaply generated H&E microscopic images of small biopsies, which are the mainstay of pathology diagnoses worldwide. The chosen innovative multi-step strategy, utilising the specific strengths of the consortium partners, will also allow the quantification and evaluation of specific microscopic features, which will improve model performance and generate insight into biological processes using a process called ‘explainable AI’, while obviating the need for expensive and complex orthogonal techniques such as sequencing or transcriptional analysis.

Due to the web-based approach in collaboration with Ellogon.AI, the technology would also be readily accessible from across the globe using universally produced H&E stains, thus democratising personalised treatment approaches for bowel cancer patients. We also anticipate a snowball effect as this project will be of interest to the surgical treatment of many other tumour types.