Reliable deep learning for coronary artery disease detection: a patient-level, statistically validated MRI study

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Diagnostics

Background:

• Accurate detection of coronary artery disease (CAD) from cardiac magnetic resonance (CMR) imaging is crucial for timely diagnosis and effective clinical management.
• Deep learning approaches offer potential for automating CAD detection, improving efficiency and accuracy.

Purpose of the Study:

• To evaluate and compare the performance and statistical robustness of DenseNet121 and ResNet50 deep learning architectures for automated CAD classification.
• To assess the reliability of these models using multiparametric CMR data.

Main Methods:

• Cardiac magnetic resonance (CMR) images were preprocessed and partitioned at the patient level.
• Model performance was quantified using average accuracy, AUC-ROC, and PR-AUC.
• Statistical tests (Shapiro-Wilk, Brown-Forsythe) assessed performance distributions and variance homogeneity.

Main Results:

• ResNet50 achieved superior performance with 90.43% accuracy, 0.862 AUC-ROC, and 0.891 PR-AUC.
• DenseNet121 demonstrated lower accuracy at 81.72%.
• Statistical analyses indicated non-normal performance distributions and significant variance differences between the models.

Conclusions:

• ResNet50 provides a reliable and statistically validated method for CAD detection from CMR imaging.
• The study highlights the importance of realistic preprocessing and comprehensive statistical testing for reproducible clinical performance estimates.
• Deep learning models, particularly ResNet50, show significant promise in advancing automated cardiac disease diagnosis.