Dual-Channel Coupling Approach Enhancing Multi-Stain Pathology Image Matching for Enhancing Cancer Diagnostics.

Abstract

Medical image matching is crucial for assisting pathological diagnosis, as it aligns gold standard hematoxylin and eosin (H&E) and immunohistochemistry (IHC) stained pathology images, enabling a comprehensive assessment for identifying cancerous regions. However, manual annotation of multi-stain pathology images incurs high labor costs. To address this challenge, we propose the deep dual-channel coupling (DDC) method for multi-stain pathology image matching. DDC utilizes virtual staining to establish two matching channels, bridging H&E-stained and IHC-stained pathology images while effectively mitigating staining variations. Subsequently, each channel undergoes guided matching using deep descriptor representations of multi-stain pathology images. Finally, a coupling strategy integrates the matching results from both channels, leveraging information from different channels to enhance accuracy and success rates. Experiment results demonstrate that DDC achieves a 93.81% success rate, surpassing the comparison method in estimating the gold standard based on 210 manual annotations. Compared to manual annotation errors, DDC improves accuracy by 45.24%, bringing it closer to the level of clinical manual annotation. Although DDC cannot replace pathologists in fully automated cancer classification, it serves as a limited aid for comprehensive assessments, demonstrating outstanding reliability in distinguishing malignant Hodgkin lymphoma and diagnosing ductal carcinoma in situ of the breast. Therefore, DDC holds significant potential in matching pathology images and supporting clinical pathological diagnostic applications.