Cardiac Region Detection Using YOLO-Based Deep Learning Models: A Performance Analysis of YOLO Models

Abstract

Cardiac region detection is a critical task for the diagnosis and management of cardiovascular diseases (CVDs), where precise identification of key areas, including the left ventricle (LV), right ventricle (RV), and myocardium (MYO), from cardiac magnetic resonance (CMR) images is essential. This study presents a comprehensive performance analysis of various YOLO (You Only Look Once) deep learning models, including YOLOv5, YOLOv8, YOLOv9, and YOLOv10, for the automatic detection of these cardiac regions. The dataset utilized comprises 2D images derived
from 3D MRI scans, segmented into training and testing sets. The models were evaluated based on standard metrics such as precision, recall, and mean average precision (mAP). Results indicate that YOLOv5l achieved the highest precision of 99.11%, while YOLOv5s recorded the best recall at 97.99%. The findings demonstrate that smaller models, such as YOLOv5n and YOLOv9t, exhibited superior precision despite their reduced computational requirements. The analysis suggests that model size does not always correlate with performance, highlighting the potential of lighter models for real-time applications in clinical settings.