Optimizing Object Detection Algorithms for Congenital Heart Diseases in Echocardiography: Exploring Bounding Box...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Congenital heart diseases (CHDs), especially septal defects, present diagnostic challenges in echocardiography.
• Current research lacks clarity on how cardiac structural information and data augmentation influence CHD detection accuracy.

Purpose of the Study:

• To evaluate the impact of incorporating cardiac structural information and data augmentation in AI models for identifying septal defects.
• To assess the performance of You Look Only Once (YOLO)v5, YOLOv7, and YOLOv9 object detection frameworks.

Main Methods:

• Utilized advanced AI object detection frameworks (YOLOv5, YOLOv7, YOLOv9).
• Investigated the effect of including cardiac structural information during the labeling process.
• Applied data augmentation techniques to echocardiographic images of septal defects.

Main Results:

• Labeling strategies significantly influence AI model performance in detecting septal defects.
• Bounding box adjustments and inclusion of cardiac structural details in annotations are critical for accuracy.
• Deep learning in echocardiography improves the precision of septal defect detection.

Conclusions:

• Meticulous annotation of medical imaging data is essential for optimizing AI object detection algorithms.
• Findings offer insights for enhancing AI applications in diagnostic cardiology.