Coronary artery disease classification using ConvMixer based classifier from CT angiography images

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) is a leading cause of death globally.
• Accurate CAD assessment is vital for treatment planning.
• Cardiac CT imaging offers high-resolution visualization but faces analysis challenges.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated CAD classification from computed tomography angiography (CTA) images.
• To investigate the efficacy of convMixer combined with morphological operations and median filters for CAD detection.

Main Methods:

• Utilized a dataset of 5,959 CTA images.
• Employed the convMixer deep learning architecture.
• Integrated morphological operations and median filters for image preprocessing.
• Classified CAD using the developed deep learning model.

Main Results:

• The convMixer model combined with morphological operations achieved 96.30% accuracy, 94.39% sensitivity, and 99.16% specificity.
• The convMixer model alone achieved 94.63% accuracy, 95.82% sensitivity, and 93.10% specificity.
• Deep learning heatmaps were used for model interpretability.

Conclusions:

• The proposed deep learning system demonstrates high accuracy in non-invasively identifying patients who may require invasive coronary angiography.
• This automated approach can enhance diagnostic efficiency, reduce manual workload, and support clinical decision-making.
• The findings support the potential for advanced AI systems in cardiovascular diagnostics and treatment planning.