Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

John M. Tyler


Losslessly compressing a medical image set with multiple slices is paramount in radiology since all the information within a medical image set is crucial for both diagnosis and treatment. This dissertation presents a novel and efficient diagnostically lossless compression scheme (predicted wavelet lossless compression method) for sets of magnetic resonance (MR) brain images, which are called 3-D MR brain images. This compression scheme provides 3-D MR brain images with the progressive and preliminary diagnosis capabilities. The spatial dependency in 3-D MR brain images is studied with histograms, entropy, correlation, and wavelet decomposition coefficients. This spatial dependency is utilized to design three kinds of predictors, i.e., intra-, inter-, and intra-and-inter-slice predictors, that use the correlation among neighboring pixels. Five integer wavelet transformations are applied to the prediction residues. It shows that the intra-slice predictor 3 using a x-pixel and a y-pixel for prediction plus the 1st-level (2, 2) interpolating integer wavelet with run-length and arithmetic coding achieves the best compression. An automated threshold based background noise removal technique is applied to remove the noise outside the diagnostic region. This preprocessing method improves the compression ratio of the proposed compression technique by approximately 1.61 times. A feature vector based approach is used to determine the representative slice with the most discernible brain structures. This representative slice is progressively encoded by a lossless embedded zerotree wavelet method. A rough version of this representative slice is gradually transmitted at an increasing bit rate so the validity of the whole set can be determined early. This feature vector based approach is also utilized to detect multiple sclerosis (MS) at an early stage. Our compression technique with the progressive and preliminary diagnosis capability is tested with simulated and real 3-D MR brain image sets. The compression improvement versus the best commonly used lossless compression method (lossless JPEG) is 41.83% for simulated 3-D MR brain image sets and 71.42% for real 3-D MR brain image sets. The accuracy of the preliminary MS diagnosis is 66.67% based on six studies with an expert radiologist's diagnosis.