Class-driven content-based medical image retrieval using hash codes of deep features

Abstract

Medical imaging provides the convenience of physicians to analyze the disease by providing visual data of the body parts required for clinical research and treatment. Today, increasing medical images following technological developments are stored for a better understanding of diseases and future diagnoses. Effective medical image indexing and retrieval systems are required to use these images from storage repositories in real-time. In this quest, this paper provides an effective indexing and retrieval framework using deep features for MR and CT image indexing and searching. The proposed system aims to produce the most effective and least parameterized hash codes by using image features. For this reason, deep features are obtained from medical images using the convolutional neural network (CNN) architecture, which is the most effective automatic feature extraction method. The length of the acquired raw deep feature vectors for an image is relatively inefficient for retrieval speed. Feature reduction methods are used for the most effective reduction of the length of the deep feature vector. The most effective feature reduction algorithm is determined in this study. The main reason for producing a reduced class-driven hash code with feature selection algorithms is the drawbacks of medical image datasets. These drawbacks prevent the CNN output from being used directly as hash-code. The performance of the proposed method is tested on NEMA MRI and NEMA CT datasets. The proposed method is able to outperform the other state-of-the-art algorithms in terms of average precision performance.