Retinopathy of Prematurity is one of the major ocular abnormalities affecting preterm infants or babies born with low birth weight. Other systemic complications also may add on to the abnormalities causing visual impairment. However, effective and earlier diagnosis is a major step in preventing the devastating complications. There are many socio-economic factors involved in the prevention of artificial blindness. One such is the availability of expertise which is linked with the imaging and diagnostic facility. Ultrasonic B Scan is commonly used to diagnose RoP. It is easily available and economical too. But it is not always suitable to identify every stage of RoP because of the overlapping of low frequency and high frequency signal components. The earlier stages of RoP appear with low clarity and severe stages of RoP appear with increased noise and echoes. Alongside, the physical errors do add up and the percentage of error increases. This may misguide the Ophthalmologist in the interpretation of images with clinical correlation. Hence, need for an improved algorithm to differentiate and identify the stages of RoP and enhance the quality of B Scan images. The conventional frequency domain algorithms like DWT, SWT may not be potential in achieving the targeted outcomes. Here, in this paper, we propose DnCNN based algorithm in order to sub-classify the retinal images, enhance the pixel quality, improve the edge detection, remove the noise components and increase the overall visibility and understanding of retinal layers, tissues and nerve fibres on the posterior end of the eye. The errors during the B Scan image acquisition are also considered to be nullified in order to improve the accuracy of interpretation of output images. On observation and clinical correlation of the result, the quality image is enhanced about 95% compared to the input RoP images. The internal structures, distribution of blood vessels, curvature, and region of abnormality are identified with improved quality and finer edges

Retinopathy of Prematurity, Ultrasonic B Scan Images, DnCNN, Image Denoising, Segmentation, Image Enhancement