Back to all blog posts
Introduction
Colorectal cancer is the third most common cancer in the United States and pathologists are becoming increasingly valuable in the management of these patients. Analysis of histological features of the tumors is the main source of data for diagnosis, but there are many components to this. Routine pathologic assessment does not capture a quantitative view of the extensive variability that colorectal cancer can demonstrate. Lymphatic, venous, and perineural invasion are some prognostic features associated with worse survival and commonly used to guide treatments. Furthermore, tumour budding (TB) and poorly differentiated clusters (PDC) are important in analysing all stages of colorectal carcinoma, particularly tumor invasiveness.
A team of pathologists developed an AI model to study novel prognostic histological features in colorectal cancer with Aiforia Create. The model was trained to segment colorectal tissue slides into 13 regions and one object. It demonstrated strong agreement with interpretations by gastrointestinal pathologists on classification of mucinous and high-grade tumors. To learn more about the experience, read our interview with Dr. Rish Pai, Consultant Pathologist and Professor of Laboratory Medicine and Pathology at the Mayo Clinic.
Interview with Dr. Rish Pai
What do you think is most unique about the project behind this publication?
Dr. Pai: This is the first publication to perform detailed segmentation of colorectal carcinoma digitized images. The quantitative output correlated well with interpretations by expert gastrointestinal pathologists.
What was the biggest benefit of the Aiforia software in this project?
The Aiforia software was very easy to use by pathologists with only limited experience in deep learning. The software is also incredibly powerful and highlights the potential of deep learning to transform the practice of pathology. The web based nature of Aiforia also facilitated collaborations among an international group of pathologists.
What are your next steps after this publication; do you plan to use AI for further research or other projects?
I plan to apply this algorithm to a large number of colorectal carcinomas to identify prognostic and predictive histologic signatures.