Medical Image Retrieval using Hybrid Features and Advanced Computational Intelligence Techniques

Abstract

In last few decades, Content-based image retrieval (CBIR) has become the most distinctive research territory in the field of computer vision. The accessibility of huge consistently developing amount of visual data and the advancement of internet emphasizes the necessity of topical access strategy that offers more than basic text-based according to user requirement. Several tools have been implemented to communicate and accomplish questions in perspective of the audio or visual material to help to filter extensive media files. Still, there is room for improvement relating to the huge databases. Digital images also play a vital role in the medical field; these images are acquired by using different equipment or different modalities. Modality identification may possibly be a standout amongst the most critical filters to confine the analysis and focus the outcomes sets and improved search results. The goal of the research is to develop a well organize classification and retrieval mechanism that can classify a given medical image on the basis of its modality. In this research, we present an approach for modality-based classification and retrieval for medical images. Experiments were made on the dataset which was used in modality classification task at imageCLEF2012 evaluation forum. Modality identification is performed by using visual features of the image. The visual features used in this research are scale-invariant feature transform (SIFT), local binary pattern (LBP), local ternary pattern (LTP), edge histogram descriptor (EHD), edge directivity descriptor (CEDD), color edge detector using wavelet transform and color histogram. Afterward, the features extracted from the image are combined to form a single feature vector. This feature vector is then given as input to the classifier to classify the image into 31 different classes.  Support vector machine (SVM) with chi-square kernel is used to solve the classification problem. Proposed framework obtained 72.2% overall accuracy. The obtained accuracy is 2.6% higher as compared to highest reported accuracy in imageCLEF2012 competition for modality classification task using visual features.