A novel perspective of ATR-FTIR spectroscopy combined with multiple machine learning methods for postmortem interval (PMI) human skin

Abstract

Due to the lack of simple, accurate, and reliable methods, the determination of PMI remains one of the most challenging tasks in forensic pathology, particularly during advanced stages of decomposition. Although numerous methods have been developed for PMI estimation, most are based on animal studies, and the extrapolation of these results to humans remains limited and questionable, providing limited practical utility. To address this gap, we collected a substantial number of human samples and focused on skin tissue, which shows significant potential but has been less extensively studied. ATR-FTIR spectroscopy combined with multiple machine learning algorithms was employed to monitor the spectral changes of skin at different PMI groups. Various algorithms (PLS-R, CLR, PCR, MLR, SVR, XGB-R, and ANN) were utilized to predict PMI. The results demonstrated that the chemical changes in lipids and proteins within postmortem skin tissue exhibited a strong time-dependent pattern. The intensity of lipid absorption peaks in fresh skin tissue was significantly higher than that in decomposed tissue, whereas amide I and II bands demonstrated the opposite trend, initially increasing and subsequently decreasing, which played a crucial role in distinguishing different time points and estimating PMI. The SVR model yielded highly satisfactory results, with the actual PMI showing close alignment with the predicted PMI. The R²CV reached 0.92, while the R²P achieved 0.96, with the RMSE as low as 2.0 days. The RMSEP/RMSECV value of 0.77 indicates the model's strong stability. These findings demonstrate that ATR-FTIR spectroscopy combined with machine learning holds significant potential and practical applicability for PMI estimation in actual forensic cases. This approach not only addresses the research gap in PMI estimation based on human skin samples but also establishes a new research direction in this field.