Multimodal physiological signal fusion with TS-CNN for early prediction of intraoperative hypotension

Intraoperative hypotension (IOH) is a common and serious complication, which is closely associated with an increased risk of postoperative organ injury and death. However, the lack of effective prospective prediction methods in clinical practice hinders early intervention to some extent. Accordingly, this study developed a TimeSeries Convolutional Network(TS-CNN) model for early IOH prediction, utilizing high-resolution physiological data from the VitalDB database. The model innovatively fuses traditional hemodynamic parameters (mean arterial pressure (MAP), systolic blood pressure (SBP)) and new physiological parameters (photoplethold pulse wave (PLETH), electrocardiogram (ECG)). The experimental results show that the four-parameter fusion model (MAP, SBP, ECG, PLETH) has the best performance in predicting IOH 10 minutes in advance, with an AUC of 0.8734, which is significantly better than the baseline model using only MAP and SBP (AUC: 0.8567). On this basis, ablation experiments further confirmed the independent predictive value of PLETH metrics and the critical positive synergistic gain when ECG is combined with PLETH. These findings verify that multimodal physiological data fusion, especially PLETH and ECG integration, can effectively achieve early IOH prediction, thereby facilitating more proactive clinical interventions to improve patient prognosis.