HATs: Hierarchical Adaptive Taxonomy Segmentation for Panoramic Pathology Image Analysis

Area of Science:

• Computational pathology
• Medical image analysis
• Artificial intelligence in medicine

Background:

• Panoramic image segmentation in computational pathology is challenging due to complex anatomy and scale variations.
• Kidney pathology involves intricate structures from regions to cellular levels, requiring precise segmentation.

Purpose of the Study:

• To propose a novel Hierarchical Adaptive Taxonomy Segmentation (HATs) method for segmenting panoramic kidney pathology images.
• To leverage anatomical insights and AI for accurate segmentation across multiple scales and object classes.

Main Methods:

• Developed the HATs technique, translating spatial relationships of 15 object classes into a "plug-and-play" loss function.
• Incorporated anatomical hierarchies and scale into a unified matrix representation.
• Utilized EfficientSAM as an AI foundation model for feature extraction, removing the need for manual prompts.

Main Results:

• HATs method demonstrated efficient and effective segmentation of kidney structures across more than 15 categories.
• The approach successfully integrated clinical insights and imaging precedents into a unified model.
• Achieved accurate segmentation by leveraging detailed anatomical knowledge and AI adaptability.

Conclusions:

• The HATs method provides a robust solution for panoramic image segmentation in kidney pathology.
• This AI-powered approach enhances the integration of anatomical and clinical data for improved pathological analysis.
• The publicly available implementation facilitates further research and application in computational pathology.