Automatic diagnosis of retinal diseases using a new method in Deep Dictionary Learning on OCT images

The aim of this research is to modify the classification efficiency of state-of-the-art techniques using a multi-layer
framework based on deep dictionary learning. This paper presents the new method in deep dictionary learning for
classification tasks and applies it to diagnose diseases of the retina. The framework uses a label consistent in the K-
SVD algorithm to learn a discriminative dictionary for sparse coding to learn better features in retinal OCT images.
Also, we used label information with each dictionary item (columns of the dictionary matrix) to enforce
discrimination in sparse codes during the multi-later dictionary learning process. This algorithm learned multiple
levels of dictionaries instead of learning shallow dictionary. The performance of the proposed algorithm is evaluated
on two datasets Duke and THOCT to diagnose diseases AMD and DME. The proposed method results are
comparable or better. The approach proposed is based on the best algorithms and is more precise than highly tuned
models and strong dictionary learning models. The proposed model of deep dictionary learning presents the idea to
develop more impressive dictionary learning methods and can help to move forward the state-of-the-art.