Optimizing Spinal Infection Classification in Whole-Slide Images via Graph Convolutional Network and Model Uncertainty Integration

Abstract

Background Spinal infections such as pyogenic spondylitis, spinal tuberculosis, and brucellar spondylitis are severe conditions that can lead to significant spinal damage and chronic pain. Whole-slide imaging (WSI) provides valuable visual information in pathological diagnoses. However, owing to the complexity and high dimensionality of WSI data, traditional manual diagnostic methods are often time-consuming and prone to errors. Therefore, developing an automated image analysis method is crucial to enhance the diagnostic accuracy and efficiency of WSI for spinal infections. Methods This study employed a novel framework that combines Graph Convolutional Networks (GCNs) with uncertainty quantification techniques to classify WSI images of spinal infections. A graph was constructed from segmented regions of the WSI, where nodes represented segmented pathological features and edges represented spatial relationships. The model was trained using a dataset of 422 cases from a provincial center for disease control and prevention and annotated for tuberculosis, brucellosis, and purulent spondylitis. The performance metrics were accuracy, precision, recall, and F1 scores. Results The integrated GCN model demonstrated a classification accuracy of 87%, recall of 85%, and F1 score of 0.86. Comparative analyses revealed that the GCN model exhibited a 10% higher performance than that of traditional CNN models. Moreover, the GCN model effectively quantified uncertainty and enhanced confidence in diagnostic decisions. Conclusions Integrating GCNs with model uncertainty enhances the accuracy and reliability of WSI image classification in pathology. This method significantly improves the capture of spatial relationships and identification of pathological features of spinal infections, offering a robust framework for supporting diagnostic and therapeutic decisions in medical practice.