# Non-invasive enhanced hypertension detection through ballistocardiograph signals with Mamba model

## TL;DR

This study uses ballistocardiography and the Mamba model to detect hypertension non-invasively, showing high accuracy for home-based monitoring.

## Contribution

The novel integration of the Mamba deep learning architecture with BCG signals for hypertension detection is introduced.

## Key findings

- The Mamba Classifier achieved 95.14% accuracy and 0.9922 AUC in hypertension detection.
- BCG signals combined with AI techniques show potential for non-invasive, long-term monitoring.
- Mamba outperformed other models like Transformer, Stacking, Voting, and XGBoost.

## Abstract

This study explores using ballistocardiography (BCG), a non-invasive cardiovascular monitoring technique, combined with advanced machine learning and deep learning models for hypertension detection. The motivation behind this research is to develop a non-invasive and efficient approach for long-term hypertension monitoring, facilitating home-based health assessments. A dataset of 128 BCG recordings has been used, capturing body micro-vibrations from cardiac activity. Various classification models, including Mamba Classifier, Transformer, Stacking, Voting, and XGBoost, were applied to differentiate hypertensive individuals from normotensive ones. In this study, integrating BCG signals with deep learning and machine learning models for hypertension detection is distinguished from previous literature by employing the Mamba deep learning architecture and Transformer-based models. Unlike conventional methods in literature, this study enables more effective analysis of time-series data with the Mamba architecture, capturing long-term signal dependencies and achieving higher accuracy rates. In particular, the combined use of Mamba architecture and the Transformer model’s signal processing capabilities represents a novel approach not previously seen in the literature. While existing studies on BCG signals typically rely on traditional machine learning algorithms, this study aims to achieve higher success rates in hypertension detection by integrating signal processing and deep learning stages. The Mamba Classifier outperformed other models, achieving an accuracy of 95.14% and an AUC of 0.9922 in the 25% hold-out validation. Transformer and Stacking models also demonstrated strong performance, while the Voting and XGBoost models showed comparatively lower results. When combined with artificial intelligence techniques, the findings indicate the potential of BCG signals in providing non-invasive, long-term hypertension detection. The results suggest that the Mamba Classifier is the most effective model for this dataset. This research underscores the potential of BCG technology for continuous home-based health monitoring, providing a feasible alternative to traditional methods. Future research should aim to validate these findings with larger datasets and explore the clinical applications of BCG for cardiovascular disease monitoring.