QROP: Quantum Learning-Based Identification of Retinopathy of Prematurity

Abstract

Retinopathy of prematurity (ROP) is a serious eye disease for premature infants. One of the main reasons for ROP is the use of oxygen for prolonged periods. In ROP, the abnormal blood vessels extend into the vitreous, the gel-like substance, and the retina may become partially detached with the formation of a ridge. Early detection and treatment of ROP are important to prevent blindness. This work aims (i) to classify normal and ROP-affected retinal images using a quantum neural network (QNN) and (ii) to compare the performance of the proposed quantum ROP (QROP) system with the existing ROP identification methods. QROP uses the HVDROPDB dataset fundus images of preterm infants. These images are captured using RetCam and Neo imaging devices. Only 15 parameters and a few samples extracted from the database were used for model training to achieve desirable evaluation metrics of accuracy, precision, sensitivity, F1-score and specificity. The proposed system achieves 97.06% accuracy with the HVDROPDB Neo dataset, 91.18% accuracy with the HVDROPDB RetCam dataset, and 85.29% accuracy when evaluated on the images from variable imaging devices and of different resolutions.