Abstract
Despite advances in non-invasive methods, endomyocardial biopsy (EMB) remains essential for definitive diagnosis of amyloidosis in many cases. Traditionally, Congo red birefringence (CRB) has been crucial for identifying amyloid deposits but is challenging to capture digitally. Emerging fluorescent Congo red imaging (CRF) overcomes this problem and holds promise in image analysis and AI applications. The diagnostic performance of CRF on virtual slides was evaluated in a cohort of EMB and autopsy cases. The feasibility of developing AI algorithms applicable to centers lacking a fluorescence scanner was investigated leveraging a computational pipeline that enables fluorescence outcome visualization in brightfield. The study analyzed 43 digital myocardial slides stained with Congo Red, acquired using a fluorescent Texas Red filter. Among these, 28 (65%) were diagnosed with amyloidosis, with complete diagnostic agreement with original diagnosis. AI achieved an AUC-ROC of 0.87, 0.86 and 0.79 on the training, validation and test set, respectively, in tile-level classification for amyloidosis positivity and IoU and Dice scores indicating partial but reasonable overlap between predictions and ground truth in amyloid segmentation. The study underscores CRF's transformative impact on virtual slides and AI integration for diagnosing cardiac amyloidosis, showcasing high reliability and diagnostic accuracy. These advancements promise a more quantitative and precise approach, facilitating the histological study of the disease in the digital transition era of pathology labs.