Deep learning exploration for SPECT MPI polar map images classification in coronary artery disease

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) diagnosis relies on myocardial perfusion imaging.
• Polar maps are a key visualization tool in myocardial perfusion studies.
• Accurate classification of polar maps is crucial for CAD detection.

Purpose of the Study:

• To explore and implement deep learning for classifying myocardial perfusion polar maps.
• To assess the effectiveness of a custom Convolutional Neural Network (CNN) and transfer learning for CAD detection.

Main Methods:

• Utilized a dataset of 144 normal and 170 pathological stress/rest polar maps (AC and NAC formats).
• Implemented transfer learning with VGG16 and data augmentation (rotation, flipping).
• Developed and evaluated a custom lightweight CNN (RGB CNN) with k-fold validation.

Main Results:

• The custom RGB CNN model achieved 92.07% agreement with a loss of 0.2519.
• Transfer learning using VGG16 attained a higher accuracy of 95.83%.

Conclusions:

• Deep learning models, particularly VGG16 transfer learning, demonstrate high accuracy in classifying myocardial perfusion polar maps.
• The proposed models offer a potential automated tool for aiding in the medical classification of CAD from perfusion imaging.