Analysis and Retrieval of Medical Images from Databases using CBIR Techniques Associated with Semantics for Diagnosis Purposes

Abstract

With enormous use of digital images in this information era, searching an newlineimage from huge collection of images has become an important necessity. Normally newlinesearching of desired images are text-based, and text-based search requires all the newlineimages to be named according to their description by a process called annotation. newlineSince the number of images is large, manual annotation becomes a very tedious newlineprocess. To eliminate this problem, accessing images based on their content, referred newlineas Content Based Image Retrieval (CBIR) came into practice. newlineThe Content-based Image Retrieval is a technique in the domain of computer newlinevision which helps to find the images in large image repositories considering its newlinecontents. Content refers to the features such as shape, texture or colour of the regions newlinein an image. Though there is a substantial improvement evidenced in the field of newlineCBIR, existing systems still are not accurate specially when applied to medical image newlinedatabases. newlineIn this work, the focus is on medical images. The Computer Aided Diagnosis newlineof Liver cancer is considered here. The Computed Tomography (CT) images of newlineabdominal region are considered. Here, first the liver portion has to be extracted from newlineabdominal CT through the segmentation technique. In this work adaptive region newlinegrowing algorithm is used to extract the liver portion. Later, the liver portion is newlinefurther segmented in order to identify tumours if any, using Linear Iterative newlineClustering technique. The tumour regions are identified and marked on segmented newlineimages. The GLCM and Histogram features of these regions are computed for newlinevalidation purpose in order to confirm that they are tumors. newlineFurther, by considering the affected portions identified and confirmed as newlinetumours, texture features are extracted by making use of Average Correction Higher newlineOrder Local Autocorrelation Coefficients (ACHLAC) and shape features are newlineextracted through Legendre Moments. Based on their features, classification of tumor newlineas benign or malignant is carried out by making use of Rough Set classifier. newlineIn this work, a type of benign tumor Haemangioma and a type of malignant newlinetumor Hepatocellular Carcinoma are considered. newline newlineii newline newlineThe abdominal CT images are taken from open repositories available in the newlinewebsites pertaining to medical domain and radiology namely radiopaedia.org and newlineradiographics.com. The work is verified by domain expert to ensure the correctness newlineand accuracy of the work. newlineThis thesis has made four significant contributions: segmentation of liver newlinefrom abdominal CT, detection of tumors from the identified liver region, newlineclassification of tumours and developing a semantic associated Content-based newlineMedical Image Retrieval (CBMIR) system. newline