الفهرس 
Representation of Medical Images
Background and Related WorkOnly 14 pages are availabe for public view
 |  | from | 135 |  |  |
| | | from | 135 | | |
Abstract
Liver segmentation from medical images has become vital to assess liver diseases and their risk level. Current techniques in imaging modalities especially MRI (Magnetic Resonance Images) has brought attention to the possibility for non-invasive diagnosis of liver diseases.
However, the challenges faced by segmentation have made it an unresolved problem so far, forcing researchers to develop several techniques to overcome these challenges. With each technique having its pros and cons; there is no universally accepted one for automatic liver segmentation.
Hence, the aim of this thesis is to try to solve this problem by developing a method of segmenting the liver from MRI images that is accurate, fast, reliable and robustness, and can be relied upon as a clinical diagnostic tool by deriving the vital information of the patient’s liver.
Two methods were proposed, with the principle of a completely different work. where the first relies on the combination of multiple efficient and reliable techniques of active contour, thresholding and some morphological processes, the second adopts the principle of probability, particularly Bayesian classifier, as it is considered the most efficient in this field. The two approaches were compared and contrasted based on accuracy and total completion time.
In the first method an initial liver segmentation is done through thresholding to obtain a liver mask that is further used as a seed to a full liver segmentation stage using the active contour model. Results are enhanced using morphological operations.
In the second proposed method, the liver is segmented directly from the MRI images by a Bayesian model without any pre-processing, and the results are also enhanced using morphological operations in the post-processing stage.Since the developed method is meant to be capable of segmenting of MRI images available in clinics, datasets were collected from several medical centers and hospitals. The datasets used for testing and training contain 20 MRI studies (400 images) obtained from different imaging modalities and comprise normal (3 cases) and diseased (17 cases) studies with different stages of liver disease progression.
The segmentation accuracy for the two proposed methods was close; an average accuracy of 95% was calculated for the active contour model, while a 95.5% for the Bayesian model.
The average processing time of liver segmentation for the active contours method was 0.96 second, while the Bayesian model takes less processing time with 0.22 second overall tested datasets.
The results are very promising to achieve the main goal of the liver segmentation process from all slices for the purpose of obtaining the liver in full size. The proposed methods are fully automatic, fast, robust and reliable as they have been tested on different medical cases collected from various magnetic resonance scanners. The accuracy of the proposed methods has also proven to be competitive with respect to previous works.