الفهرس Only 14 pages are availabe for public view
 |  | from | 152 |  |  |
| | | from | 152 | | |
Abstract
It is very common for radiologists, looking at a medical image (MI) to make quick circles around any abnormality for diagnosis or segmentation purposes. However automatic segmentation of region of interest (ROI) is an increasing research field for both compression and computer aided diagnosis. This thesis introduces modified methods for biomedical image processing, telemedicine, and archiving. The proposed automatic segmentation method uses an introduced artificial neural network (IANN) and introduces a modified fuzzy technique; whereas in coding it uses a modified arithmetic encoder. Moreover, based on the important observation that large wavelet coefficients are much more important than small wavelet coefficients, a proposed fast and reduced bit embedded zerotree wavelet (FEZW) algorithm is introduced. This FEZW aims to reduce the transmission time of images in low bandwidth environments, and to reduce storage space as well.
The IANN is used to determine the expected center of the ROI area, whereas the fuzzy technique is used for critical points determination to obtain an automatic ROI contour. The first proposed fuzzy system produces 64 fuzzy rules. To reduce the number of rules three steps have been suggested. First an exclusive rule is defined, then difference fuzzy concept is introduced, and finally inverted rules reduction is applied. These rules reduction in turns reduce fuzzy system complexity. The introduced automatic ROI segmentation is compared with the manual and semi- automatic techniques, that had several disadvantages and discouraged physicians to have confidence in semi- automatic segmentation. This introduced difference fuzzy model (IDFM) had better solving segmentation problems, when applied to several types of medical images with different shaped and different intensity variations inside and outside the ROI.
After being automatically segmented, the ROI is coded lossless using a modified arithmetic coding (MAC) method. This MAC surpasses the conventional arithmetic coding (AC) lossless compression technique in reducing computational complexity and storage size. The IDFM combined with the MAC form a hybrid lossless compression channel. The coded ROI will be added to the rest of the image (coded lossy using the FEZW) through a proposed bi- channel automatic compression technique (PBACT). This PBACT consists of two channels: a lossy channel that uses the FEZW and th~ hybrid lossless compression channel. This PBACT aims to obtain suitable compression ratios and good image qualities.
Later, the hybrid lossless compression channel is then modified for better compression ratios without losing the best image quality. This modified hybrid compression channel (MHCC) uses the FEZW with a higher refinement level inside the ROI to give higher compression ratios than the MAC lossless technique. Thus, only one coding method inside and outside the ROI is applied, which gives simplicity and reliability to the modified bi- channel.
Several magnetic resonance brain images and fluorescene ophthalmic images are applied to the proposed bi- channel technique. Simulation results are presented to validate the proposed and the introduced techniques.
Generally, the thesis is organized in nine chapters as follows:
Chapter 1 is an introduction for medical imaging and problems facing their techniques.
Chapter 2 is concerned with wavelet based MI processing. In this chapter, a survey of the wavelet transform (WT) and its use in MI coding is shown, then it explains the reasons for selecting the EZW to process MIs as one of the most promising coding techniques for ~ompression and progressive image transmission.
Chapter 3 is a survey of different MI segmentation and compression methods. Semi- automatic segmentation techniques have received much attention in this chapter, whereas selective image compression is described in full details.
Chapter 4 introduces the fast and reduced bit embedded zerotree wavelet (FEZW) algorithm, as a method for compression and progressive transmission of MIs.
Chapter 5 presents the proposed neuro- fuzzy automatic segmentation method for automatic determination of the ROI.
Chapter 6 firstly introduces a new concept called difference fuzzy (DF) method. Second, it applies this proposal as an automatic segmentation method to determine ROI of different images. Finally, it presents the results of using this introduced difference fuzzy model (IDFM) and shows how it better solves the difficult
problems ofMI segmentation.
Chapter 7 introduces a proposed hybrid lossless compression channel. Simulation results show how the modified arithmetic coding (MAC) supports and reduces computational complexity and storage size.
Chapter 8 introduces two proposed bi- channel automatic compression techniques (PBACT) for selective compression of MI. The first bi-channel uses the hybrid loss less compression channel with the MAC for the ROI coding; whereas the second is based on a modified hybrid c0mpression channel (MHCC) that uses the FEZW with higher refinement levei inside the ROI than outside.
Chapter 9 presents the main conclusion derived from the introduced proposals, and suggests other problems for future study.