Automated identification and quantification of myocardial inflammatory infiltration in whole slide image to diagnose myocarditis

Computational pathology provides new opportunities for the rapid diagnosis of myocarditis using whole-slide images (WSIs). However, the high complexity and fine-grained features of myocarditis histopathological images limit the specificity of existing methods. To address this, we propose an automated identification and quantitative analysis method that leverages myocarditis histopathological features to detect myocardial inflammatory infiltration. This method employs deep learning techniques for cell nucleus segmentation, selecting lymphocyte nuclei based on their characteristics. The aggregation state is assessed by calculating the distances between lymphocyte nuclei, which enables the identification of inflammatory infiltration regions. The lymphocyte nuclei within these regions are further counted, and the average lymphocyte nucleus density (LND) of the WSI is calculated. The LND threshold is optimized using three-fold cross-validation for accurate myocarditis diagnosis. Our method achieved an accuracy of 0.905, sensitivity of 0.895, and an AUC of 0.915 on the test set. The specificity reached 0.936, representing a significant improvement of 29.1% over the clustering-constrained attention multiple instance learning (CLAM) network. Additionally, experiments demonstrate that the method performs effectively in identifying inflammatory infiltration in both myocarditis and cardiac immune rejection WSIs, with all accuracies exceeding 0.97.