An algorithm for cardiac disease detection based on the magnetic resonance imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Cardiac magnetic resonance imaging (MRI) is crucial for diagnosing heart disease.
• Existing object detection models struggle with accuracy and reliability in cardiac MRI analysis.
• There is a need for advanced methods to improve automated detection of cardiac pathologies.

Purpose of the Study:

• To propose SA-YOLO, an innovative object detection method for enhanced cardiac MRI analysis.
• To address the limitations of current models in detecting heart disease from medical images.
• To improve the accuracy and reliability of object detection in cardiac MRI.

Main Methods:

• Developed SA-YOLO, a modified YOLOv8 model for cardiac MRI.
• Replaced Spatial Pyramid Pooling Fast with Multi-Channel Spatial Pyramid Pooling.
• Integrated a novel attention mechanism (Squeeze-Excitation and Coordinate Attention) into the model's Neck.
• Utilized iSD-IoU loss, combining shape and distance loss, for bounding box regression.

Main Results:

• SA-YOLO demonstrated superior performance in detecting cardiac pathologies on the Automated Cardiac Diagnosis Challenge dataset.
• Achieved a 7.4% improvement in mAP0.5 and a 5.1% improvement in mAP0.5-0.95 compared to the baseline YOLOv8 model.
• The proposed modifications led to more accurate and reliable object detection in cardiac MRI.

Conclusions:

• SA-YOLO represents a significant advancement in automated cardiac MRI analysis.
• The enhanced model offers improved accuracy for heart disease detection.
• This method holds promise for clinical applications in cardiology and medical imaging diagnostics.