Neurological Disorder Detection Using OCT Scan Image of Eye

Extracting useful features from medical images (in Radiology) helps to improve the early precautions and treatment as well as reducing the risk of performing major surgeries. There is an immense need of a methodology which can automatically extract retinal features and classify it into desired class. The proposed automated technique is based on Deep Convolutional Neural Networks of four different types of Convolutional Neural Networks (CNN) models i.e. Baseline 5 layer model, AlexNet and ResNET. These three CNN techniques are used for image segmentations, features extraction and classification of normal and abnormal images in three different image formats: Choroidal Neovascularization (CNV), DRUSEN (small yellow colored deposits of debris of eyes, and Diabetic Macular Edema (DME). Furthermore, the optimizers used are adaptive moment estimation (ADM) and stochastic gradient descent (SGD) which provide different accuracy under same iterations. ADM optimizer gives the best accuracy in all the CNN networks while SGD gives the least accuracy. Various type of abnormalities are detected in retina images are CNV, DME, and DRUSEN. The method proposed in this research is implemented on the Optical coherence tomography (OCT) Retinal Image dataset. For the sake of validation of proposed methodology, some known performance parameters like accuracy, F1_score, recall, precision, support and loss function are used. The observed and visualized results of the proposed method Compared to previous techniques, the findings of the approach are encouraging.