A novel contrast enhancement technique for diabetic retinal image pre-processing and classification

• Published in: International ophthalmology Vol. 45; no. 1; p. 11
• Main Authors: Naz, Huma, Ahuja, Neelu Jyothi
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 16.12.2024; Springer Nature B.V
• Subjects: Algorithms; Blindness; Classification; Clustering; Color imagery; Comparative analysis; Diabetes; Diabetes mellitus; Diabetic retinopathy; Diabetic Retinopathy - classification; Diabetic Retinopathy - diagnosis; Diabetic Retinopathy - diagnostic imaging; Diagnosis; Female; Fundus Oculi; Humans; Image Enhancement - methods; Image filters; Image Processing, Computer-Assisted - methods; Machine learning; Male; Medicine; Medicine & Public Health; Microaneurysm - diagnosis; Middle Aged; Noise intensity; Noise reduction; Ophthalmology; Original Paper; Retina - diagnostic imaging; Retina - pathology; Retinal images; Retinal Vessels - diagnostic imaging; Retinal Vessels - pathology; Retinopathy; Support vector machines; Diabetic retinopathy pre-processing; Diabetic retinal image pre-processing; Retinal fundus image; Contrast enhancement
• ISSN: 1573-2630; 0165-5701; 1573-2630
• DOI: 10.1007/s10792-024-03377-2

Background Diabetic Retinopathy (DR) is a leading cause of blindness among individuals aged 18 to 65 with diabetes, affecting 35–60% of this population, according to the International Diabetes Federation. Early diagnosis is critical for preventing vision loss, yet processing raw fundus images using machine learning faces significant challenges, particularly in accurately identifying microaneurysm lesions, which are crucial for diagnosis. Methods This study proposes a novel pre-processing technique utilizing the Modified Fuzzy C-means Clustering approach combined with a Support Vector Machine classifier. The method includes converting RGB images to HSI colour space, applying median filtering to reduce noise, enhancing contrast through Intensity Histogram Equalization, and identifying false microaneurysm candidates using connected components. Additionally, morphological operations are performed to remove the optic disc from the enhanced images due to its similarity to microaneurysms. Results The proposed method was evaluated using publicly available datasets, demonstrating superior performance compared to existing state-of-the-art algorithms. The approach achieved an accuracy rate of 99.31%, significantly improving the detection of microaneurysms and reducing false detections. Conclusions The findings indicate that the proposed pre-processing technique effectively enhances diabetic retinopathy classification by addressing the challenges of false microaneurysm detection. The comparative analysis against state-of-the-art algorithms highlights the effectiveness of the proposed method, particularly in addressing the challenges associated with false microaneurysms.