Preprocessing retinal fundus images to localize lesions for identification of diabetic eye diseases

Diabetic eye diseases (DiED) are the foremost difficulty for diabetic patients and may lead to visual impairment. Early identification followed by appropriate therapy planning eventually supports to minimize the risk. In literature, various significant state-of-the-art AI enabled diagnostic models h...

Full description

Saved in:
Bibliographic Details
Main Authors: Das, Pallabi, Nayak, Rajashree
Format: Conference Proceeding
Language:English
Subjects:
Blood vessels
Diagnostic systems
Eye diseases
Fatalities
Homogeneity
Image enhancement
Image processing
Image quality
Image segmentation
Medical imaging
Quality assurance
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Diabetic eye diseases (DiED) are the foremost difficulty for diabetic patients and may lead to visual impairment. Early identification followed by appropriate therapy planning eventually supports to minimize the risk. In literature, various significant state-of-the-art AI enabled diagnostic models have been suggested to provide intelligent early identification of DiED by analyzing rich amount of high-quality retinal fundus images (RFIs). RFIs with limited sharpness, improper illumination and non-uniform homogeneity etc, may expedites to misdiagnosis. Hence, quality of RFIs should be efficiently enhanced before feeding to any diagnostic model. This paper suggests a systematic image processing based pre-processing framework to boost up the quality of RFIs for further processing. The proposed framework performs noise reduction, image enhancement, retinal vessel segmentation and subsequently executes transfer learning-based classification process to identify DiEDs. Efficacy of the quality assured RFIs in the diagnosis of DiED is established through comprehensive experiments.
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0166533