الفهرس 
INTODUCTIONOnly 14 pages are availabe for public view
 |  | from | 126 |  |  |
| | | from | 126 | | |
Abstract
Advances in information technology and telecommunications have acted as catalysts for significant developments in the sector of health care. This technological advance has had a particularly strong impact in the field of medical imaging (Mr), where film radiographic techniques are gradually being replaced by digital imaging techniques, and this has provided an impetus to the development of integrated hospital information systems which support the digital transmission, storage, retrieval, analysis, and interpretation of distributed multimedia jJatient records.
One of the most important MI systems is computer aided diagnosis (CAD). CAD is access to high-performance computing facilities in order to execute computationally intensive MI analysis and have successfully revealed masses and microcalcifications on screening MI with highly added-value services in breast cancer field.
Medical image denoising is an essential preprocessing step not only for medical CAD systems, but also for image processing field. Image denoising still remains a challenge for researchers, because noise removal introduces artifacts and causes blurring of the images.
The wavelet transform has become an important tool to suppress the noise, due to its effectiveness and properties. The functionality of denoising in wavelet domain, called wavelet shrinkage, is due to threshold value and the thresholding function. Many wavelet transform based noise reduction algorithms will be presented and its assumptions, advantages, and limitations will also be discussed.
The main problem in the wavelet shrinkage is the selection of an optimal threshold value. Thresholding Neural Network (TNN) is used to seek the optimal threshold value; using a Gradient-based learning algorithms to perform adaptive image denoising. Some modifications are proposed to improve the TNN capability with remarkable time improvement and good image quality. The proposed denoising method uses, as an application, noisy tumor MIs of breast.
Segmentation is used as a second phase in to extract an accurate Region Of Interest (Ra I) after noise removal step.
This thesis introduces a semi-automatic thresholding based method for extracting the ROI from tumor breast MI. Simulation results are presented to validate the proposed modified TNN method. Results are observed and compared with relevant existing work.