Improved YOLOv5s-based lesion area detection method for ophthalmic ultrasound images
10.19745/j.1003-8868.2024206
- VernacularTitle:基于改进YOLOv5s的眼科超声影像病灶区域检测方法研究
- Author:
You ZHOU
1
;
Ze-Meng LI
;
Xin-Qi YU
;
Xiao-Chun WANG
;
Sheng ZHOU
Author Information
1. 中国医学科学院北京协和医学院生物医学工程研究所,天津 300192
- Keywords:
YOLOv5s;
ophthalmic ultrasound;
ultrasound image;
deep learning;
ophthalmic disease;
lesion area detection
- From:
Chinese Medical Equipment Journal
2024;45(11):1-7
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose an improved YOLOv5s-based lesion area detection method for ophthalmic ultrasound images so as to solve the problems due to high complexity,difficult deployment and low accuracy of the model during ophthalmic ultrasound imaging detection and diagnosis.Methods Firstly,an ophthalmic ultrasound image dataset was established contai-ning Lhe images of stellate vitreous degeneration,retinal detachment,vitreous hemorrhage,posterior vitreous detachment and posterior scleral staphyloma.Secondly,a YOLOv5s-MobileNetV2 model was constructed based on YOLOv5s with the original backbone feature extraction network CSPDarkNet replaced by the lightweight network MobileNet.Thirdly,the model's performance in recognizing lesion areas in ophthalmic ultrasound images was evaluated by multi-category mean average precision(mAP),number of parameters and frames per second(FPS).Finally,the intelligent detection software for ophthalmic ultrasound images was designed based on PyQt5 library.Results The YOLOv5s-MobileNetV2 model had the mAP,number of parameters and FPS being 97.73％,4.61×106 and 47 f/s respectively,which gained advantages in timeliness over YOLOv5s by decreasing the mAP by 0.22％and the number of parameters by 34.98％.The developed intelligent detection software for ophthalmic ultrasound images behaved in human-computer interaction and clinical applicability of YOLOv5s-MobileNetV2 model.Conclusion The improved YOLOv5s-based lesion area detection method for ophthalmic ultrasound images meets clinical diagnosis requirements for ophthalmic diseases by involving in lightweight models and detecting lesion areas accurately.[Chinese Medical Equipment Journal,2024,45(11):1-7]
