Deep neural network model for hypertension risk level detection using ECG signal

• Main Author: Lau, Kai Yun
• Other Authors: Vidya Sudarshan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Medicine, Health and Life Sciences; Hypertension risk level; Neural network; ECG signal
• Online Access: https://hdl.handle.net/10356/175003
• Institution: Nanyang Technological University

Background and Objectives: Early detection of hypertension risk is crucial as it affects about 1 in 4 adults and accounts for about half of all heart diseases and stroke-related deaths globally. Traditionally, assessments can be time-consuming and prone to human error. This study proposes an Electrocardiogram (ECG) based model for hypertension risk level detection. Methods: Hypertensive ECG recordings from SHAREE and Normotensive ECG recordings from PTB-XL were utilized. These ECG recordings undergo Daubechies-6 wavelet transform and a median filter to denoise and prepare the data for a deep convolutional network. The deep convolutional network consists of 1D convolution and max pooling layers, which extract feature mappings and perform risk level classifications. It classifies the ECGs into ‘normotensive’ (no risk), ‘hypertension’ (low risk), and ‘hypertension + any recorded event within 12 months of recording’ (high risk). Results: The multi-class classification achieved a micro-average accuracy of 99.59%, indicating excellent overall performance. Individual classification precision ranged from 97.8% to 99.74% while recall exceeds 97.31% for all categories. The high accuracy and strong individual classification performances demonstrate the effectiveness of the proposed model in accurately identifying hypertension risk levels. Conclusion: This study demonstrates the potential of the proposed deep neural model for hypertension risk level detection. However, a more clinical approach would still be required to validate the model’s reliability.