4.6 Article

Detection and classification of red lesions from retinal images for diabetic retinopathy detection using deep learning models

期刊

出版社

SPRINGER
DOI: 10.1007/s11042-023-15045-1

关键词

Diabetic retinopathy; Red lesions; Microaneurysms; Haemorrhages; CNN; UNet

向作者/读者索取更多资源

Diabetic retinopathy is an eye disease caused by diabetes mellitus that can lead to vision loss if not treated early. Manual diagnosis requires physical tests, which are time-consuming and costly. This study aims to develop an automated model using retinal images to detect early-stage diabetic retinopathy based on red lesions.
Diabetic retinopathy (DR) is an eye disease caused by retinal damage induced by the long-term illness of diabetes mellitus. In the early stages, DR may show no symptoms or only minor vision difficulties, but it can eventually result in vision loss if not treated early. Manual diagnosis of Diabetic retinopathy requires many physical tests like visual acuity, pupil dilation, and tonometry. But the issue with these tests is that they consume more time, cost, and effort, therefore affecting patients. Also, it is challenging to identify the disease through these tests during the earlier stage of the disease due to its tiny structure. Since the current system is highly dependent on the resources available and time-consuming, better alternatives are being sought. The proposed work aims to develop an automated model for detecting the early stage of DR detection based on red lesions present in the retinal images. The pre-processing is carried out to remove the noise present, improve local contrast levels in the image, and is further subjected to UNet architecture for the semantic segmentation of red lesions. Advanced Convolutional Layer Architecture in U-Net was used to support pixel-level class labeling, which is much needed in medical segmentation. The segmented images obtained from the red lesion detection were then used as the input to feed the convolution neural network to train and classify the input images to their corresponding severity classes. Four publicly available datasets, IDRiD, DIARETDB1, MESSIDOR, and STARE, were used in the proposed model to evaluate its performance. On working with the IDRID dataset, the specificity and sensitivity of the proposed Red Lesion detection system were observed as 99% and 89%, respectively, with an accuracy of 95.65%. The specificity, sensitivity, and accuracy obtained for the Diabetic retinopathy severity classification system were 93.8%, 92.3%, and 94%, respectively, on working with the MESSIDOR dataset.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.6
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据