PathEX: Make good choice for whole slide image extraction

Area of Science:

• Digital Pathology
• Computational Pathology
• Medical Image Analysis

Background:

• Whole slide image (WSI) analysis often uses tile-based approaches for slide-level predictions.
• Irregular tumor shapes and dimensions present challenges in WSI analysis.
• Existing methods struggle with accurate tile extraction around complex tumor boundaries.

Purpose of the Study:

• To introduce PathEX, a novel framework for precise tile image extraction in WSI analysis.
• To address the limitations of current methods in handling irregular tumor regions.
• To improve the efficiency and accuracy of tile selection for pathology model training.

Main Methods:

• Developed PathEX, integrating intersection over tile (IoT) and background over tile (BoT) algorithms.
• Utilized PathEX for tile extraction around annotated region boundaries, excluding blank tiles.
• Trained classification models on CAM and PAIP datasets using eight generated training sets with varied IoT and BoT parameters.

Main Results:

• PathEX successfully extracts tile images at annotated region boundaries, with adjustable extraction based on IoT parameter.
• Exclusion of blank tiles within annotated regions is achieved by adjusting the BoT parameter.
• Optimal performance was observed with specific IoT and BoT parameter ranges, indicating the framework's adaptability.

Conclusions:

• PathEX offers a robust solution for tile image extraction in digital pathology, integrating IoT and BoT algorithms.
• The framework allows researchers to customize tile extraction by setting IoT and BoT thresholds.
• PathEX provides valuable guidance for optimizing tile image selection in WSI analysis for improved diagnostic accuracy.