1DCNN-BiLSTM-transformer hypertension risk prediction model based on APW.

Introduction: Hypertension has a multifactorial etiology. Recent studies have revealed a link between hypertension and gut microbiota dysbiosis. Pulse wave analysis holds significant clinical value for hypertension risk assessment. While research on deep learning models utilizing photoplethysmography (PPG) for hypertension classification has advanced, limitations persist. PPG offers limited richness and accuracy for characterizing blood pressure-related pathological information. In contrast, Arterial Pressure Waveform (APW) provides richer pathological information and exhibit stronger correlations with clinically interpretable features. However, deep learning research using APW for hypertension classification remains limited, as existing studies focus primarily on local feature extraction and neglect global temporal dynamics. Methods: To address these challenges, we propose a novel 1D-CNN-BiLSTM-Transformer architecture for hypertension risk assessment based on APW, where the 1D-CNN module extracts waveform morphology features from signals within individual pressure segments, the BiLSTM module models long-range temporal dependencies from signals within each segment, and the Transformer module explicitly captures nonlinear interaction from signals across different pressure segments through multi-head self-attention mechanisms. Results: We use the multi-channel APW database from the Population Health Data Archive (PHDA), containing hypertensive and non-hypertensive cases with APW signals acquired from six traditional Chinese medicine points (left-cun, left-guan, left-chi, right-cun, right-guan, and right-chi) to evaluate the model's performance. The model outperforms the current state-of-the-art methods in accuracy, precision, recall, and F1 score across all six points. Conclusion: The proposed model enhances classification performance. The physiologically driven interpretable analysis demonstrates that APW can reflect pathophysiological features associated with gut microbiota dysbiosis. The model-driven interpretable analysis offers a decision-making basis for clinical diagnosis.