To develop a deep learning framework for the automated detection of intestinal metaplasia in gastric histopathology images.
Approach:
Model Framework: The CNXTGeM model integrates a ConvNeXt-Tiny backbone with Generalized Mean pooling and Efficient Channel Attention to enhance feature representation.
Data Evaluation: The model was evaluated using 1,037 H&E-stained gastric biopsy samples and externally validated with the GasHisSDB dataset.
Model Interpretability: Interpretability was assessed using gradient-based visualization techniques: Grad-CAM, Grad-CAM++, and XGrad-CAM.
Key Findings:
CNXTGeM achieved an accuracy of 99.04%, precision of 98.08%, specificity of 98.11%, and an F1-score of 99.03%.
The model demonstrated 100% sensitivity, improving recall by 8.51% over the baseline ConvNeXt model.
On the external GasHisSDB dataset, CNXTGeM maintained an accuracy of 99.34% and an F1-score of 99.31%.
Interpretation:
Limitations:
The study's findings are based on a specific dataset and may require further validation across diverse populations.
Potential challenges in AI implementation in clinical settings were not fully addressed.
