U-Net Segmentation of Ultra-Widefield Retinal Fundus Images for Retinal Vein Occlusion Associated Lesion Recognition
Inspection of fundus images can identify haem-orrhages and cotton wool spots associated with Retinal Vein Occlusion (RVO) disease. Detection of the lesion in fundus images using a computer can aid in early interventions. Previous studies have employed image-processing techniques and feature engineer...
保存先:
| 主要な著者: | , , , |
|---|---|
| フォーマット: | Proceeding |
| 言語: | English |
| 出版事項: |
2023
|
| 主題: | |
| オンライン･アクセス: | http://ir.unimas.my/id/eprint/43426/3/U-Net.pdf http://ir.unimas.my/id/eprint/43426/ https://ieeexplore.ieee.org/document/10291722 |
| タグ: |
タグ追加
タグなし, このレコードへの初めてのタグを付けませんか!
|
| 要約: | Inspection of fundus images can identify haem-orrhages and cotton wool spots associated with Retinal Vein Occlusion (RVO) disease. Detection of the lesion in fundus images using a computer can aid in early interventions. Previous studies have employed image-processing techniques and feature engineering approaches to identify features from regular images and construct machine learning models. There are limited studies on using Ultra-widefield (UWF) fundus images for machine learning owing to the lack of gold-standard datasets. In addition, it investigates the efficacy of prediction models trained with regular images in predicting RVO symptoms in UWF images. This study employed a deep learning approach to detect RVO lesions and subsequently used them to construct a classifier. We leveraged regular fundus images for lesion segmentation due to the limited availability of public UWF datasets and compared their effectiveness with a segmentation model trained solely on UWF images. Our results found that the segmentation model trained on regular fundus images is less effective in detecting haemorrhages and cotton wool spots in UWF images. Finally, we found that the lesion regions work perfectly in building a classifier that can discriminate between RVO and non-RVO fundus images. |
|---|