Deep Learning-Based Detection Approach for Uterine Lymph Node MRI Images: A Pilot Study
Juping Qiu, Chunxia Chen, Ming Li, Yongping Lin · 2024
Automatic detection of lymph nodes (LN) in uterine magnetic resonance imaging (MRI) images not only helps radiologists process images quickly and accurately, but also effectively assists clinical diagnosis and improves diagnostic accuracy. However, it is challenging to develop an effective detection model due to the uneven size distribution of MRI images of uterine LN, as well as issues such as missing or blurred borders and significant morphological differences. In this study, we conducted a retrospective study using 158 MRI images. Five detection models (SSD, Faster R-CNN, CenterNet, YOLOv8, and YOLOv10) were trained in our collection of MRI images. The experimental results demonstrated that the YOLOv8 model outperformed the YOLOv10 model, achieving the best detection performance in the test dataset. At an Intersection over the Union (IoU) threshold of 0.50, the model reached an Average Precision (AP) of 0.75 and an F1 score of 0.65. At the IoU threshold of 0.25, the AP was 0.81 and the overall F1 score was 0.76. The detection results provide practical references for doctors who diagnose LN MRI images.