Abstract Introduction Cardiac allograft rejection (CAR) is a major cause of graft failure, and accurate detection is crucial for timely intervention. Current histopathological evaluation of endomyocardial biopsy (EMB) samples relies on manual inspection, which is time-consuming and subject to interobserver variability. Deep learning approach leveraging digital pathology may improve the accuracy and consistency of rejection diagnosis. In recent years, deep learning has been increasingly explored as a tool to assist in the detection of rejection, demonstrating promising results in automating the identification of lymphocyte infiltrations. However, further research is needed to refine these approaches and improve their clinical applicability. Purpose This study aims to develop a deep learning-based approach for detecting lymphocyte infiltration in EMB to improve the diagnosis of cardiac allograft rejection. By leveraging digital pathology, we seek to enhance diagnostic accuracy and reduce interobserver variability. Additionally, we aim to demonstrate the effectiveness of AI in this task, even when working with a limited amount of data, highlighting the potential for AI-driven solutions in resource-constrained settings. Methods We analyzed 702 histological slides from 171 patients, all scanned manually to ensure high-quality digitization. To extract meaningful features from samples, we employed UNI, a foundational model specifically designed for digital pathology. These features were processed using an Attention-based Multiple Instance Learning (MIL) pooling approach, which aggregates information across multiple slides per patient and provides a more robust rejection diagnosis. The dataset was split into 70% for training and 30% for testing to evaluate model generalization. To assess performance, we used AUC-ROC and accuracy as our primary metrics. Results On the test set, our model achieved an average AUC-ROC of 0.88 and an average accuracy of 0.78. Notably, the recall score exceeds 0.99 in certain tests, highlighting a strong sensitivity to detecting positive cases. However, the system shows some specificity limitations, as biopsy sites can exhibit patterns resembling lymphocytic infiltrates associated with rejection, occasionally leading to false positives. Conclusion Our results demonstrate the potential of deep learning in improving the diagnosis of cardiac allograft rejection by enabling an automated and consistent assessment of lymphocyte infiltration in EMB samples. By leveraging UNI for feature extraction and an Attention-based MIL pooling approach, we achieved strong performance despite the limited amount of available data. These results suggest that while the model effectively captures lymphocytic infiltration patterns; further refinements are needed to improve specificity and overall diagnostic reliability.
Buccieri et al. (Sat,) studied this question.