ColoPath-Mamba: a stain-aware pathology foundation–Mamba framework for six-class colorectal histopathological image classification
Year: 2026
Type: article
Abstract: Colorectal histopathological image classification plays a critical role in computer-aided diagnosis. However, discrimination between visually similar tissue types is often difficult due to the presence of various subtypes of glands, epithelial abnormalities, and staining patterns in hematoxylin and eosin (H&E) stained images. This study proposes ColoPath-Mamba, a stain-aware pathology foundation–Mamba framework for six-class colorectal histopathological image classification using the EBHI-Seg dataset. The proposed framework first applies stain-aware preprocessing and then extracts pathology-oriented representations using a pretrained pathology foundation encoder. A task-specific pathology adaptation layer is introduced to align the extracted features with EBHI-Seg diagnostic categories. To capture long-range tissue dependencies, a Vision Mamba-based tissue-context block is incorporated, followed by class-aware attention pooling for diagnostic feature aggregation. In addition, a stain-consistency regularization strategy is used to reduce sensitivity to controlled stain perturbations by enforcing representation similarity between stain-normalized and stain-augmented views of the same image. The model classifies images into six diagnostic categories. Experimental evaluation using a five-fold protocol shows that ColoPath-Mamba achieves 97.60% accuracy, 98.28% weighted precision and 97.72% weighted F1-score. Comparative and ablation analyses demonstrate that pathology foundation features, Mamba-based tissue-context modelling, class-aware attention, and stain-consistency learning jointly improve classification performance.
Source: Scientific Reports
Authors: Amit Kumar(Conceptualization, Methodology, Writing – original draft), Harishchander Anandaram(Conceptualization, Supervision), Nimisha Gupta(Formal analysis, Methodology), Pinki Sagar(Formal analysis), Ganesh Shridhar Raghtate(Writing – review & editing), Sankalp Paliwal(Writing – original draft, Writing – review & editing)
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236147410
