DeepLabV3+Ensemble for Diagnosis of Cardiac Transplant Rejection

Heart transplantation is a complex procedure, often joined with complications such as cardiac transplant rejection. Current diagnostic methods include regular invasive and time-consuming biopsies followed by histopathological analysis. Deep learning has the potential to significantly enhance speed and objectivity and introduce new information from the obtained sample to increase the chances of predicting rejection. Our study presents several deep-learning approaches for quantitative analysis of histological scans for acquiring supportive information. The proposed segmentation methods focus on inflammation, endocardium, and blood vessels. The study compares the experimental results of multiple methods evaluated using real data from medical experts. This study lays the groundwork for future research and demonstrates the potential of deep learning applied to the prediction of transplant rejection.