MSAI-Path: Predicting Microsatellite Instability From Routine Histology Slides Without Reinventing the Wheel.

Area of Science:

• Computational pathology
• Biomarker discovery
• Colorectal cancer research

Background:

• Microsatellite instability (MSI) is a critical biomarker in colorectal cancer, impacting patient outcomes and treatment strategies.
• Current deep learning models for MSI prediction from whole-slide images (WSI) lack interpretability, hindering clinical trust and validation.
• Pathologist expertise in identifying histologic features for MSI assessment is not fully integrated into computational models.

Purpose of the Study:

• To develop a novel, explainable, and verifiable hybrid approach for MSI prediction in colorectal cancer.
• To combine computational analysis of histologic features with pathologist expertise for improved MSI prediction.
• To create a trustworthy alternative to black-box deep learning models in computational pathology.

Main Methods:

• A hybrid method was developed using nuclei and tissue segmentation to quantify MSI-associated histologic features (e.g., intraepithelial lymphocytes, differentiation grade, mucinous components, tertiary lymphoid structures) based on Bethesda guidelines.
• These computationally derived features were integrated with clinical data and analyzed using logistic regression and random forest models for MSI status prediction.
• The approach was validated on 3256 WSI from 2267 patients across multiple cohorts and centers.

Main Results:

• The hybrid method achieved an area under the curve (AUC) of up to 0.88 for resections and 0.90 for biopsies, comparable to black-box deep learning models.
• Learned variable importances strongly correlated with manual scoring systems and pathologist assessments, ensuring interpretability.
• The approach demonstrated potential as a screening tool, capable of excluding 41% of patients from gold-standard MSI testing with 95% sensitivity.

Conclusions:

• Classifiers based on clinical and validated histologic information can predict MSI status as effectively as black-box models while providing complete interpretability.
• This explainable approach offers a trustworthy pathway for biomarker prediction in computational pathology.
• The method holds promise for streamlining MSI testing and improving diagnostic efficiency in colorectal cancer.