Forensic cause-of-death determination of myocardial infarction-related sudden cardiac death in histopathologically similar cases using computational pathology.

Determining whether myocardial infarction-related sudden cardiac death (MI-SCD) is the cause of death remains particularly challenging in forensic practice because different causes of death can produce highly similar myocardial histopathological changes. This study developed a computational pathology framework to support forensic cause-of-death determination by distinguishing MI-SCD from other causes associated with similar myocardial alterations and identifying characteristic myocardial injury patterns associated with MI-SCD. The study included 412H&E-stained whole-slide images (WSIs) of left ventricular myocardium from the GTEx database (221 MI-SCD and 191 control cases) and an independent external validation cohort of 43 slides (18 MI-SCD and 25 control cases). Image features were extracted using the UNI encoder, and slide-level classification was performed using a clustering-constrained attention multiple-instance learning model (CLAM). The UNI-CLAM model achieved area under the curve (AUC) values of 0.895, 0.908, and 0.829 and accuracies of 0.815, 0.824, and 0.790 in the test, validation, and external cohorts, respectively. Beyond classification, attention-guided analysis indicated that the model integrated histopathological information from multiple tissue patterns rather than relying on any single lesion. Subsequent spatial co-localization analysis identified characteristic myocardial injury patterns in MI-SCD, including the frequent co-occurrence of myocardial rupture and degeneration and the close spatial association of myocardial fibrosis with both degeneration and inflammatory cell infiltration. These findings suggest that MI-SCD is characterized by distinctive myocardial injury patterns despite substantial histopathological overlap with other causes of death and may provide a more objective basis for forensic cause-of-death determination.