# Enhancing WSI image classification with graph convolutional neural networks and model uncertainty modeling

## TL;DR

This paper introduces a new method combining graph neural networks and uncertainty modeling to improve the accuracy of classifying whole slide images in pathology.

## Contribution

The novel integration of graph convolutional networks with uncertainty quantification for WSI classification is presented.

## Key findings

- The GCN model achieved 87% accuracy in classifying spinal infections from WSI images.
- The model outperformed traditional CNNs by 10% in classification performance metrics.
- Uncertainty quantification improved confidence in diagnostic decisions.

## Abstract

The primary research question addresses whether integrating Graph Convolutional Neural Networks with model uncertainty modeling can improve the accuracy and robustness of Whole Slide Imaging (WSI) classifications in pathology.

This study employed a novel framework combining GCNs with uncertainty quantification techniques to classify WSI images of spinal infections. We constructed a graph from segmented regions of WSI, where nodes represented segmented pathological features and edges represented spatial relationships. The model was trained on a dataset of 422 cases from the Shandong Provincial Center for Disease Control and Prevention, annotated for tuberculosis, brucellosis, and purulent spondylitis. Performance metrics included accuracy, precision, recall, and F1 score.

The integrated GCN model demonstrated a classification accuracy of 87%, with a recall of 85% and an F1 score of 0.86. These metrics signify an improvement over traditional CNN models, which showed a 10% lower performance in comparative analyses. The model also effectively quantified uncertainty, enhancing confidence in diagnostic decisions.

Integrating GCNs with model uncertainty modeling enhances the accuracy and reliability of WSI image classification in pathology. This approach significantly improves the capture of spatial relationships and pathological feature recognition, offering a robust framework for supporting diagnostic and therapeutic decisions in medical practice.

The enhanced ability to classify and understand WSI images using this method has significant implications for pathology, potentially leading to more accurate and reliable diagnoses. This approach could be particularly useful in remote diagnostics and in environments where expert pathological consultation is limited.