Title: An investigation of retinal descriptors on Indian database for automatic pathological diagnosis and classification of retinopathy of prematurity

Authors: Sushma S. Kadge; Sanjay L. Nalbalwar; Anil B. Nandgaonkar; Parag K. Shah; V. Narendran

Addresses: Electronics and Telecommunication Engineering Department, D.B.A.T. University, Lonere, Raigad, India; K.J. Somaiya College of Engineering, Vidyavihar, Mumbai, India ' Electronics and Telecommunication Engineering Department, D.B.A.T. University, Lonere, Raigad, India ' Electronics and Telecommunication Engineering Department, D.B.A.T. University, Lonere, Raigad, India ' Department of Pediatric Retina and Ocular Oncology, Aravind Eye Hospital, Coimbatore, India ' Department of Pediatric Retina and Ocular Oncology, Aravind Eye Hospital, Coimbatore, India

Abstract: Early diagnosis is crucial to prevent blindness in preterm neonates. Scarcity of specialists indicates an urgent need for automated identification, classification and diagnosis of retinopathy of prematurity (ROP). Previous automation works did not consider the classification of ROP stages which is an essential decision maker in treatment. We developed an automatic assessment system for ROP classification (AASRC) on Indian databases. We studied stochastic gradient descent (SGD) along with five other classifiers. 64 experiments were conducted to explore the intra database based on grey-level co-occurrence matrix (GLCM) descriptors using various frequency-based parameters for classification of ROP. Information gain (IG) scoring function is used to identify best descriptor while student's t-tests is used for validation. The classification accuracy rate of ROP disorder achieved are 99.03%, 93.87%, 94.55%, 92.51% and 97.95% respectively for normal/abnormal, stages 1, 2, 3, and 4. The experimental findings demonstrate the proposed feature descriptors and classifier outperforms state-of-the-art models.

Keywords: retinopathy of prematurity; ROP; retina image analysis; classification; stochastic gradient descent; SGD; grey-level co-occurrence matrix; GLCM.

DOI: 10.1504/IJBET.2023.132887

International Journal of Biomedical Engineering and Technology, 2023 Vol.42 No.4, pp.398 - 422

Received: 20 Oct 2021
Received in revised form: 28 Jul 2022
Accepted: 15 Aug 2022
Published online: 14 Aug 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article