Mirafra Technologies | Top ASIC VLSI SOC Semiconductor Design Services Company | RTL Design, Verification, UVM, Gate Level Simulation, STA, Physical Design, Signoff, Analog Layout, Embedded Software, Enterprise Application, IOT, M2M, Data Analytics, Mobile, Networking, Automotive, Storage, Communications, Infotainment, Linux, Android, RTOS, Kernel, BSP, Porting, Device, Driver, Automotive, Embedded, IoT, Cloud, Analytics, Bigdata, Testing, Networking, Storage, Service

Automatic Retinal Disease Classification

World Health Organization (WHO) in their recent study in March 2018 has classified a few eye conditions like diabetic retinopathy (an early stage of DME -Diabetic Macular Edema), glaucoma and AMD (age-related macular degeneration) that can lead to visual impairment which can be treated properly with early diagnosis. For example, Diabetic retinopathy is the fastest growing cause of blindness, with nearly 415 million diabetic patients at risk worldwide. Medically a specialist detects such eye diseases after examining lot of pictures of the eye to determine the presence and severity of such disease, and our paper here tries to address this through machine learning techniques in an automated manner to assist such a specialist. In this research study, we present unique approaches for each of the following: automatic disease classification, pixel level clustering for both the diseases and region marking of the affected retinal areas in the OCT (Optical Coherence Tomography) image. The image is first classified and then used for segmentation. Classification model building involves data pre-processing where we have implemented a novel algorithm for curvature flattening of the images, finding Region of Interest (ROI), cropping ROI, extracting the feature, training and then evaluating the classifier. We have implemented two different models using Support Vector Machine (SVM) and Convolution Neural Network (CNN). Our study shows CNN outperforms SVM to predict disease class. Our segmentation approach focuses mainly on the area between RNFL (Retinal Nerve Fiber Layer) and RPE (Retinal Pigment Epithelium) layer of the retina. We also implemented a novel algorithm for pixel-level clustering and region marking in the images based on the symptom captured in OCT images for both DME and AMD. Our classification model has been validated using test data and we were able to achieve 100% accuracy in predicting the disease case. We have calculated average dice score to calculate the performance of region marking and achieved reasonable accuracy. This predictive system has a high potential to assist doctor in early diagnosis of AMD and DME patients