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

Area of Science:

• Cardiovascular medicine
• Biomedical engineering
• Artificial intelligence in healthcare

Background:

• Hypertension is multifactorial, with emerging links to gut microbiota dysbiosis.
• Photoplethysmography (PPG) has limitations in capturing detailed blood pressure pathology for hypertension classification.
• Arterial Pressure Waveform (APW) offers richer pathological information but has limited deep learning research.

Purpose of the Study:

• To develop a novel deep learning architecture for hypertension risk assessment using APW.
• To overcome limitations of existing methods in extracting comprehensive temporal and morphological features from APW signals.
• To enhance the accuracy and interpretability of hypertension classification based on APW.

Main Methods:

• Proposed a 1D-CNN-BiLSTM-Transformer architecture for APW analysis.
• 1D-CNN extracts local waveform morphology; BiLSTM models long-range temporal dependencies.
• Transformer captures nonlinear interactions across segments via multi-head self-attention.

Main Results:

• Evaluated on a multi-channel APW database from PHDA, including hypertensive and non-hypertensive cases.
• The proposed model significantly outperformed state-of-the-art methods in accuracy, precision, recall, and F1 score.
• Performance was consistent across APW signals from six traditional Chinese medicine points.

Conclusions:

• The novel deep learning model significantly enhances hypertension classification performance using APW.
• Physiologically driven analysis confirms APW reflects pathophysiological features linked to gut microbiota dysbiosis.
• Model-driven interpretability provides a basis for clinical hypertension diagnosis.