Artificial Intelligence-Based Tool for Tumor Detection and Quantitative Tissue Analysis in Colorectal Specimens.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Digital diagnostics

Background:

• Digital pathology enables computational algorithms for routine pathology tasks.
• Accurate tissue segmentation and tumor detection are crucial for colorectal cancer diagnosis and treatment.

Purpose of the Study:

• To develop and clinically validate a clinical-grade artificial intelligence (AI) tool for precise multiclass tissue segmentation in colorectal specimens.
• To validate the AI tool for tumor detection in colorectal biopsy specimens.

Main Methods:

• Trained a semantic segmentation algorithm on 241 manually annotated whole-slide images (WSIs) for 11 tissue classes.
• Developed an additional module for biopsy WSI classification.
• Validated the AI tool on six external case cohorts from five pathology departments across four countries using diverse scanning systems.

Main Results:

• Achieved a composite multiclass Dice score of up to 0.895 for tissue segmentation.
• Demonstrated pixel-wise tumor detection specificity and sensitivity of up to 0.958 and 0.987, respectively.
• Clinical validation showed AI tool sensitivity of 1.0 and specificity up to 0.969 for tumor detection in biopsy WSIs, with analysis time under 1 minute per case.

Conclusions:

• Developed and validated a highly accurate, clinical-grade AI tool for assistive diagnostic processing of colorectal specimens.
• The tool enables quantitative deciphering of colorectal cancer tissue for biomarker development and personalized oncologic care.
• Open-sourced annotated datasets contribute significantly to the colorectal cancer computational pathology field.