Repeated Structuring & Learning Procedure for Detection of Myocardial Ischemia: a Robustness Analysis

Area of Science:

• Cardiology
• Artificial Intelligence
• Machine Learning

Background:

• Myocardial ischemia, a reduction in heart blood flow, poses risks including sudden cardiac death and arrhythmias.
• Timely identification of myocardial ischemia is critical for patient outcomes.
• The Repeated Structuring and Learning Procedure (RS&LP) is an algorithm for dynamically creating neural networks (NNs) for disease detection.

Purpose of the Study:

• To perform a robustness analysis of the RS&LP algorithm for myocardial ischemia detection.
• To evaluate the impact of varying key parameters (NL, NI, NC) on NN performance.
• To establish the reliability of RS&LP for clinical myocardial ischemia detection.

Main Methods:

• Utilized 13 serial ECG features from 84 myocardial ischemia cases and 398 controls.
• Trained and tested NNs using 50% of the data each, with varying maximal number of layers (NL), initializations (NI), and confirmations (NC).
• Compared NN performance using the area under the curve (AUC) of receiver operating characteristics.

Main Results:

• 12 out of 13 (92%) developed NNs achieved an AUC of ≥80%.
• 4 out of 13 (31%) NNs achieved an AUC of ≥85%.
• The RS&LP algorithm demonstrated consistent performance across different parameter values.

Conclusions:

• The RS&LP algorithm is robust for developing NNs for myocardial ischemia detection.
• The study validates the reliability of RS&LP for clinical applications in identifying myocardial ischemia.
• The findings support the use of RS&LP in developing accurate diagnostic tools for cardiac conditions.