Biomedical Image Analysis for the Detection of Malignancies in Liver

Abstract

Biomedical images - microscopic and macroscopic - are analyzed for the diagnosis of malignancies as well as to know the progress in treatment. Microscopic image analysis consider microscopically small objects such as blood cells, body tissues etc.. and on the other hand macroscopic image analysis consider images of internal organs such as heart, liver, lung, pancreas etc.. newlineIn this thesis macroscopic image analysis of liver is performed for the detection of malignancies from CT images. newlineEven though several research works were published for the early detection of liver tumor using noninvasive techniques, fool proof technologies being used in clinical applications are very few in numbers. Therefore more attention needed in the preprocessing and post processing of the images, to retrieve more information. newlineAn automatic diagnosis system to classify liver diseases is addressed here. Major challenges faced in this work are segmentation of liver area and liver tumor, proper representation and measurement of tumor, optimum feature extraction and selection of proper classifier to classify the tumor. newlineThus an automatic diagnosis system is proposed in this work to classify the tumor into four categories such as Normal, Hepato Cellular Carcinoma (HCC), Hepatic Adenoma and Hemangioma. The challenge of segmentation of liver area and tumor from liver area are addressed with simple, reliable and faster image processing techniques. The possibility of data compression using wavelet transform and representation of segmented tumor are also addressed. Performance of these types of segmentation techniques is compared. A feature extraction technique named, quotHybrid Texture Descriptive Feature (HTDF)quot, vector is implemented here to extract optimum feature vector from the segmented image. The selection of feature parameters to form feature vector is by mimicking human diagnosis. The performance of HTDF feature vector is compared with two other feature extraction techniques namely SFTA and modified SFTA. newline