Application of Neural Networks in Cardiovascular Load Analysis

Abstract

This study investigates deep learning models for estimating aerobic (AeT) and anaerobic (AnT) thresholds using heart rate variability (HRV) analysis. Two CNN-LSTM architectures were developed: one predicting AeT and AnT values directly and another using signal delineation for enhanced threshold identification. The models were trained on HRV data from 119 subjects performing treadmill or cycle ergometer tests, with DFA alpha 1 used for threshold estimation. Performance evaluation showed an MAE of 4.67 bpm for AeT and 4.70 bpm for AnT in the first model, while the second model achieved 6.47 bpm for AeT and 3.15 bpm for AnT. Both models outperformed traditional DFA a1-based methods, with the second model demonstrating greater consistency in AnT detection. These results highlight the potential of deep learning for non-invasive endurance training optimization and cardiovascular monitoring.