Plus disease classification in Retinopathy of Prematurity using transform based features

Retinopathy of prematurity (ROP) is a leading cause of childhood blindness affecting the retina of low birth weight preterm infants. Plus disease in ROP characterised by abnormal tortuosity and dilation of posterior retinal blood vessels, is a benchmark that identifies treatment-requiring ROP cases. A Plus disease classifier with zero false negatives is a major requirement of an ROP screening system. In this paper, an efficient Artificial Neural Network (ANN) architecture with an optimal feature set is proposed which meets the above requirement. A total of 178 images with 81(45%) Plus and 97 (55%) No Plus are used for the analysis. A feature set derived from transform domain representation of retinal funds images is used along with the existing vascular features in the proposed work. Wavelet and Curvelet transforms are considered for deriving the additional feature set in the experimental analysis. The feature set containing Curvelet transform energy coefficient along with the vascular features gave an Accuracy of 96% and Specificity of 93% with 100% Sensitivity.