Deep Learning for Pathology: YOLOv8 with EigenCAM for Reliable Colorectal Cancer Diagnostics

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Deep learning for medical imaging

Background:

• Colorectal cancer (CRC) is a leading cause of cancer mortality worldwide.
• Accurate and timely diagnosis is crucial for effective CRC treatment.
• Current diagnostic methods like manual histopathology face observer variability, and computational tools often lack interpretability.

Purpose of the Study:

• To develop and validate a deep learning framework for accurate and interpretable colorectal cancer lesion classification.
• To integrate the YOLOv8 architecture with EigenCAM for transparent AI explanations in histopathology.
• To establish a clinically dependable foundation for AI-assisted CRC diagnosis.

Main Methods:

• A dataset of 5000 H&E-stained colorectal tissue slides was acquired and preprocessed.
• Five YOLOv8 variants were comparatively evaluated for multiclass lesion classification.
• EigenCAM was employed for visualizing discriminative regions, enhancing model interpretability.
• Statistical validation methods including Bland-Altman plots and CDFs were used to assess robustness.

Main Results:

• The YOLOv8 XLarge model achieved 99.38% training accuracy and 96.62% testing accuracy.
• The framework demonstrated superior performance compared to existing CNN- and Transformer-based systems.
• EigenCAM visualizations successfully highlighted key regions driving the AI's predictions.
• Extensive statistical validation confirmed the framework's reliability and robustness.

Conclusions:

• The developed deep learning framework offers a precise and interpretable solution for AI-assisted colorectal cancer diagnosis.
• This approach addresses limitations of manual assessment and current computational methods by combining high accuracy with visual explanations.
• The framework represents a significant advancement towards the clinical deployment of AI in pathology workflows for improved CRC detection.