CONTENT BASED MEDICAL IMAGE RETRIEVAL: A DEEP LEARNING APPROACH

Abstract

Ethiopia is one of the countries where overall health service has been compromised by inadequate & poorly maintained infrastructure and scarcity of health professionals. Radiological service is a resource intensive unit in a hospital and most developing countries radiological service is expected to be poor or may not be available at all. However, there is no study conducted to assess the radiological service in Ethiopia. Content-based medical image retrieval systems are designed to retrieve images that are relevant, based on detailed analysis of latent image characteristics. A Content-based medical image retrieval system maintains high-level image visuals in the form of feature vectors, which the retrieval engine leverages for similarity based matching and ranking for a given query image. In this study, a Content-based medical image retrieval system is proposed for the retrieval of medical images for enabling the early classification of different type of diseases based on X-ray images. The Content-based medical image retrieval system is built using the deep learning models for the retrieval and classification of disease specific features using transfer learning based like VGG16, VGG19 and ResNet50. The models have been trained on standard X-ray image datasets. The dataset contains 4194 X-ray images. From this, 80% of the images are used for training and the rest for testing the model. In this research work the distance of each query image measure by Euclidean distance, content based image retrieval based on medical database. Experimental evaluation on the standard dataset revealed that the proposed approach achieved an accuracy of 96.74% for VGG16, accuracy of 96.46% for VGG19 and accuracy of 92.30% for ResNet50. Accordingly, VGG16 is proposed based on its performance. In this study there is no means to propose medicine for the disease, proper therapy for the disease, or there is no estimate the severity of the disease once it has been classified on the medical X-ray images, which are left as a way forward for further study.