Deep learning based tissue analysis predicts outcome in colorectal cancer

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence
• Medical Imaging

Background:

• Machine learning, particularly deep learning, demonstrates expert-level accuracy in medical image classification.
• Accurate prediction of colorectal cancer patient outcomes is crucial for effective treatment strategies.

Purpose of the Study:

• To develop and evaluate a deep learning model for direct prediction of colorectal cancer patient outcomes using histopathological images.
• To compare the prognostic performance of the deep learning model against human expert histological assessment.

Main Methods:

• A deep convolutional and recurrent neural network architecture was trained on digitized hematoxylin-eosin-stained tumor tissue microarray (TMA) samples from 420 colorectal cancer patients.
• The model directly predicted patient outcome without intermediate tissue classification.
• Performance was evaluated by comparing the model's ability to stratify patients into low- and high-risk groups against expert visual assessment on TMA spots and whole slides.

Main Results:

• The deep learning model achieved superior performance in predicting patient outcomes (Hazard Ratio [HR] 2.3, Area Under the Curve [AUC] 0.69) compared to human expert assessment (HR 1.67-1.65, AUC 0.58-0.57).
• The model accurately stratified patients into low- and high-risk groups.
• The deep learning approach extracted more prognostic information from tissue morphology than experienced human observers.

Conclusions:

• State-of-the-art deep learning techniques can effectively predict colorectal cancer patient outcomes directly from histopathological images.
• Deep learning models offer a more accurate and objective prognostic tool than traditional histological assessment by human experts.
• This approach holds potential for improving risk stratification and guiding treatment decisions in colorectal cancer management.