Related Experiment Video
Updated: Oct 20, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Weakly supervised anomaly segmentation in retinal OCT images using an adversarial learning approach.
Jing Wang1,2, Wanyue Li1,2, Yiwei Chen2
1Jiangsu Key Laboratory of Medical Optics, Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Science, Suzhou 215163, China.
Biomedical Optics Express
|September 13, 2021
Summary
This study introduces a weakly supervised deep learning model for retinal lesion segmentation in optical coherence tomography (OCT) images. The novel approach accurately detects lesions in real-time without manual labeling, accelerating diagnosis.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Manual lesion segmentation in medical images is time-consuming and requires expertise.
- Supervised deep learning models necessitate extensive manual labeling and may miss novel lesions.
Purpose of the Study:
- To develop a weakly supervised learning network for accurate lesion segmentation in optical coherence tomography (OCT) images.
- To address the limitations of manual segmentation and supervised deep learning in lesion detection.
Main Methods:
- A CycleGAN-based weakly supervised learning network was proposed for lesion segmentation in full-width OCT images.
- The model reconstructs normal anatomy from abnormal images; lesions are identified by image differences.
- A customized architecture and multi-scale perceptual loss were employed to handle shape deformations.
Main Results:
- The model achieved 96.94% AUC for image-level anomaly detection and a 0.8239 Dice similarity coefficient for pixel-level lesion detection.
- Performance surpassed all comparative methods on an open-source retinal OCT dataset.
- Average test time per image was 0.039 seconds, demonstrating real-time capability.
Conclusions:
- The proposed weakly supervised model accurately detects and segments retinopathy lesions in real-time without manual labeling.
- This method has the potential to expedite clinical diagnosis and reduce misdiagnosis rates in ophthalmology.
