Enhancing glioma immunohistochemical image classification through color deconvolution-aware prior guidance

Abstract

Glioma diagnosis and prognosis heavily rely on immunohistochemistry (IHC), particularly CD34-stained images which highlight tumor vascular endothelial cells. However, traditional image analysis methods struggle with complex staining patterns and subtle morphological variations across glioma subtypes. In this study, we propose a novel Prior-Guided Enhancement Network (PGE-Net) that integrates domain-specific prior knowledge through color deconvolution to enhance feature representation of CD34-positive regions. Unlike existing approaches that treat all pixels equally, our model leverages color abnormality maps to emphasize diagnostically relevant staining patterns, thereby improving both interpretability and classification performance. Experimental evaluation on a curated glioma CD34 dataset demonstrates that PGE-Net achieves notable improvements over ResNet18 baselines, with Precision, Recall, and F1-score increased by 9.17%, 9.35%, and 12.35%, respectively. These results underscore the model’s potential for facilitating more accurate and interpretable IHC image analysis in clinical practice, ultimately supporting more personalized and efficient glioma treatment planning.