Manjunatha Mahadevappa - Digital Health Symposium and Roundtable

Abstract: Optical coherence tomography (OCT) is a standardized diagnostic imaging modality that generates cross-sectional images of the retina to diagnose retinal diseases. This work presents both classical image processing techniques for image enhancement and modern machine learning algorithms for the classification of retinal diseases. Using image processing techniques, this work represents the problem of speckle noise in OCT images. An adaptive noise reduction filter is developed, which is invariant to the amount of noise present in the image [1]. The designed filter successfully reduced the noise in the image while preserving edges [1]. Two methods based on Artificial Intelligence (AI) techniques is developed. In the first method, a joint Self-organizing map (SOM) and a K-mean clustering approach to separate the retinal layers from the background is presented [2]. Since the technique is unsupervised, it does not warrant a labelled dataset. In the second method, AI-based work, have been developed, a Convolution Neural Network architecture to classify retina diseases [3]. In this proposed work, a residual block focusing more on the retinal boundaries by enhancing the local contrast and a custom activation function for the dense layers is demonstrated [3].