Paper Details

Abstract

Cataracts, the clouding of the eye’s lens that leads to progressive vision impairment, remain a major global health challenge. Early and accurate diagnosis is critical for preventing disease progression and improving quality of life, yet traditional grading methods depend on ophthalmologists and specialized equipment, which can be costly and inaccessible in many regions. To address this, we proposed a computer-assisted diagnostic framework for cataract detection and severity classification using fundus retinal images. The approach integrated pretrained Inception and Xception networks for feature extraction, principal component analysis for feature fusion and dimensionality reduction, and ensemble classification using CNN, ANN, and MLP models with majority voting across four severity levels: normal, mild, moderate, and severe. Evaluated on 1,600 fundus images (augmented to 3,000) annotated by an expert ophthalmologist, the framework achieved 98.74\% accuracy, outperforming existing methods. These results highlighted its potential as a reliable, affordable, and accessible tool for early cataract screening and diagnosis.

Keywords
cataract feature fusion deep learning CNN fundus images
Contact Information
Minh Anh Hoang (Corresponding Author)
Computer Science Department, Swinburne University Vietnam, FPT University, HCM City, Vietnam, Vietnam
0944657599

All Authors (2)

Minh Anh Hoang C

Affiliation: Computer Science Department, Swinburne University Vietnam, FPT University, HCM City, Vietnam

Country: Vietnam

Email: minhha10@fe.edu.vn

Phone: 0944657599

Khuong Nguyen-Vinh

Affiliation: School of Science, Engineering and Technology, RMIT University, HCM City, Vietnam

Country: Vietnam

Email: khuong.nguyenvinh@rmit.edu.vn

Phone: 0935505753