Noninvasive diagnosis of ischemia non-obstructive coronary using machine learning-assisted magnetocardiography

Background The clinical diagnosis of Ischemia with Non-Obstructive Coronary Arteries (INOCA) is a cumbersome and complex task. Magnetocardiography (MCG), a highly sensitive, non-contact, and non-invasive technique, is recognized for its ability to detect subtle electrophysiological changes induced by myocardial ischemia.The aim of this study is to utilize the characteristic of MCG to achieve accurate and rapid diagnosis of INOCA. Methods The study enrolled 703 patients who underwent coronary angiography. Among them, 550 exhibited slow coronary flow without subepicardial coronary stenosis, while 203 were lesion-free. MCG was performed in all patients prior to angiography. A hierarchical feature extraction framework with ensemble learning was implemented for MCG signal analysis. QRST segmentation algorithms established unified computational baselines. Feature extraction comprised statistical parameters and temporal analysis utilizing cubic spline interpolation and wavelet transforms for time-frequency domain characterization. Feature selection employed a cascade approach: variance inflation factor analysis, ElasticNet regularization (α=0.5), and BORUTA iterations (n=500). XGBoost modified with Focal Loss (γ=2.0) was selected after comparative evaluation, with hyperparameters optimized via OPTUNA. Clinical variables were integrated using a robust architecture tolerant to missing data. Model performance was evaluated through stratified five-fold cross-validation. Results The intelligent diagnostic model achieved an average accuracy of 0.785, sensitivity of 0.770, and specificity of 0.797 across five-fold validation. The area under the ROC curve (AUC) reached 0.832, demonstrating good discriminative ability. The model effectively incorporated clinical prior knowledge by embedding basic clinical information (age, gender, diabetes history, typical symptoms) into the feature matrix, enhancing its practical utility in clinical settings Conclusion This study confirmed the diagnostic efficacy of MCG in patients with INOCA. It successfully constructs a rapid clinical diagnosis model based on machine learning, thereby providing an effective tool for the clinical diagnosis of INOCA.