Identification of Pathological Factors Related to Cardiovascular Disease Based on Pathological Image Analysis

Abstract

Cardiovascular disease ranks among the leading causes of death in Japan, highlighting the need to enhance both prevention and treatment strategies. Carotid endarterectomy (CEA), a surgical procedure that removes plaques accumulated in the carotid artery, helps reduce the risk of stroke. However, key aspects of long-term prognosis and postoperative complications following CEA remain unclear. This study aimed to identify pathological factors associated with the modified Rankin Scale (mRS) score by analyzing histopathological images of excised plaques. We used an autoencoder (AE) to extract feature representations from the pathological images and applied Lasso regression to explore their relationship with mRS scores. The analysis yielded a coefficient of determination (R2) of 0.84, indicating that the AE-derived features retain information related to the mRS score. Since the analysis was conducted only on the training data, evaluating the generalization performance remains a task for future work. We also visualized image regions associated with the mRS score using the trained regression model.