IUM-hybrid model for enhanced CAD diagnosis using deep learning and VS Grad-CAM visualization

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) is a leading global health concern.
• Early CAD detection is crucial for preventing myocardial infarction and sudden death.
• Existing models like InceptionV3, MobileNetV2, and U-NetR have limitations in feature extraction, dataset requirements, and computational efficiency for CAD classification.

Purpose of the Study:

• To develop an enhanced CAD classification model by merging InceptionV3, U-NetR, and MobileNetV2.
• To overcome the limitations of individual models in processing complex medical image data.
• To improve the accuracy and interpretability of CAD detection.

Main Methods:

• A hybrid model (IUM) was created by integrating pre-trained InceptionV3, U-NetR, and MobileNetV2 architectures.
• The model was fine-tuned on an angiographic dataset.
• Dynamic weighting was employed to maximize prediction accuracy.
• VS Grad-CAM visualization was utilized for model interpretability.

Main Results:

• The IUM model achieved high diagnostic performance metrics.
• Achieved accuracy of 0.97, F1-score of 0.99, specificity of 0.98, and sensitivity of 0.97.
• VS Grad-CAM provided precise heatmaps for elucidating classifier decisions.

Conclusions:

• The novel hybrid approach significantly enhances diagnostic precision for CAD.
• The model offers a scalable and efficient solution for clinical applications, minimizing manual errors.
• This method has the potential to improve patient outcomes through prompt and accurate CAD identification.