Related Experiment Video
Updated: Jun 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Automatic exudate and aneurysm segmentation in OCT images using UNET++ and hyperreflective-foci feature based bagged
Rinrada Tanthanathewin1, Warissaporn Wongrattanapipat1, Tin Tin Khaing1
1School of Information, Computer and Communication Technology, Sirindhorn International Institute of Technology, Thammasat University, Meung, Patumthani, Thailand.
Plos One
|May 24, 2024
Summary
This study introduces an efficient algorithm for automatically segmenting exudates and aneurysms in optical coherence tomography (OCT) images, improving diabetic retinopathy (DR) diagnosis. The method achieves high accuracy, outperforming existing techniques.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Diabetic retinopathy (DR) signs like exudates (EXs) and aneurysms (ANs) originate beneath the retinal surface, detectable via optical coherence tomography (OCT).
- Early detection of EXs and ANs aids in timely DR diagnosis, but their segmentation in OCT images is challenging due to small size, image noise, and low contrast.
- Limited public datasets and unsatisfactory performance of existing automatic segmentation methods hinder progress in DR detection.
Purpose of the Study:
- To propose an efficient algorithm for automatic segmentation of exudates (EXs) and aneurysms (ANs) in OCT images.
- To improve the accuracy and reliability of detecting early signs of diabetic retinopathy (DR).
Main Methods:
- A deep-learning U-Net++ model was employed to identify potential areas of EXs and ANs.
- Adaptive thresholding segmented candidate EX-AN regions, followed by feature extraction (appearance, location, shadow markers).
- Bagged tree ensemble classifiers were trained using these features to refine segmentation and isolate EX-AN blobs.
Main Results:
- The proposed algorithm achieved an average recall of 87.9%, precision of 86.1%, and F1-measure of 87.0% on a public dataset of 80 OCT images.
- The F1-measure significantly surpassed comparative methods: binary thresholding and watershed (BT-WS) by 78.0% and adaptive thresholding with shadow tracking (AT-ST) by 82.1%.
Conclusions:
- The developed algorithm offers an efficient and accurate solution for segmenting EXs and ANs in OCT images.
- This method holds significant potential for improving automated diabetic retinopathy screening and diagnosis.
