Non-contrast multimodal cardiac MRI for predicting coronary microvascular dysfunction in patients with hypertrophic cardiomyopathy.
Authors
Affiliations (5)
Affiliations (5)
- School of Medical Imaging, Binzhou Medical University, Guanhai Street No. 346, Laishan District, Yantai 264003, Shandong Province, China.
- Department of Radiology, Zibo Central Hospital, No.10, Shanghai Street, Zibo 255000, Shandong Province, China.
- Department of Radiology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, No.324, Jingwu Street, Jinan 250021, Shandong Province, China.
- School of Medical Imaging, Shandong Second Medical University, No. 4948, East Shengli Street,Weifang 261000, Shandong Province, China.
- Department of Radiology, Zibo Central Hospital, No.10, Shanghai Street, Zibo 255000, Shandong Province, China. Electronic address: [email protected].
Abstract
In hypertrophic cardiomyopathy (HCM), detection of coronary microcirculatory dysfunction (CMD) usually relies on contrast-enhanced cardiac magnetic resonance (CMR). This study sought to develop a practical non-contrast radiomics model to identify CMD, minimizing reliance on contrast agents. A total of 290 patients with HCM were stratified by the presence or absence of CMD and randomly allocated into a training set and a test set at an 8:2 ratio. The application of logistic regression was implemented to identify predictive imaging features. Radiomics features were extracted from the end-diastolic four-chamber view of the left ventricle and the end-diastolic short-axis view with maximal wall thickness across cine, T1 mapping, and T2 fat-saturation images. Five distinct machine learning algorithms were then employed to construct radiomics models, and ensemble models were generated by integrating features from different imaging planes. Model performance was evaluated in the test set using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA). Random forest (RF) outperformed other machine learning algorithms. Nine predictive models were constructed: S1, F1, and SF1 (from cine images); S2, F2, and SF2 (from T1 mapping); and S3, F3, and SF3 (from T2-weighted fat-saturation images), along with ensemble models. Among them, the SF2 model showed the best diagnostic performance in the test set, achieving an AUC of 0.90, accuracy of 0.83, sensitivity of 0.87, specificity of 0.75, and an F1 score of 0.87 for detecting coronary microcirculatory dysfunction. Calibration and decision curve analyses further demonstrated that SF2 was well-calibrated and offered superior clinical utility. The SF2 radiomics model, integrating T1 mapping features, demonstrated the best diagnostic performance for detecting CMD in HCM patients. These findings indicate that non-contrast radiomics holds promise as a potential alternative to contrast-enhanced CMR, with the capacity to reduce reliance on contrast agents in CMD assessment.