High-Resolution Histopathological Image Classification Model Based on Fused Heterogeneous Networks with...

Area of Science:

• Digital Pathology
• Computational Biology
• Medical Image Analysis

Background:

• Pathological diagnosis relies on analyzing whole slide images (WSIs), which presents challenges due to their high resolution.
• Machine learning (ML) offers potential for automated WSI analysis, enhancing diagnostic efficiency, objectivity, and consistency.

Purpose of the Study:

• To propose a novel ML model for the classification of WSIs.
• To address the challenge of processing high-resolution WSIs with deep neural networks.

Main Methods:

• A two-part model was developed: a self-supervised encoding network (UNet-like) to create compressed latent representations (feature cube) from WSI patches, preserving location information.
• A classification network fused four heterogeneous network blocks, taking the feature cube as input to integrate diverse features.

Main Results:

• The proposed model effectively encodes WSI features while preserving spatial information.
• The fused network integrated heterogeneous features, leading to robust classification performance.
• Evaluation on two public datasets demonstrated the model's effectiveness compared to baseline approaches.

Conclusions:

• The novel model demonstrates a promising approach for automated WSI classification.
• The method effectively handles high-resolution image data, paving the way for improved pathological diagnosis.
• Integrating heterogeneous features enhances the robustness and accuracy of ML-based diagnostic tools.