الفهرس 
Chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 151 |  |  |
| | | from | 151 | | |
Abstract
This thesis is concerned with the processing of medical images to extract useful
information from these images. A general framework is adopted in this thesis that
comprises degradation reduction, segmentation and classification. This framework
can be considered as a step towards automatic diagnosis based on medical images.
Different types of medical images are considered in this thesis including Ultrasonic
(Us), X-ray (XR), Computed Tomography (CT), Positron Emission Tomography
(PET) and Magnetic Resonance (MR)images. The degradation effect in each type of
images are considered. Different noise reduction algorithms are developed to reduce
the effect of degradation on the images prior to segmentation. The segmentation
process of images depends on the proper selection of the segmentation algorithm. The
proposed segmentation algorithms comprise two trends. The first one depends on the
improved and fast fuzzy C-means (IFFCM) with the particle swarm optimization
(PSO). The second one depends on the fuzzy C-means (FCM) with morphological
operations and active contour segmentation. The last step in the proposed framework
is the classification of segmented tumors to classify them as being benign or
malignant. Both statistical and CNN classifiers are considered for the tumor
classification process. Moreover, to cope with emergency case of coronavirus disease
(COVID-19) spread, we developed an efficient classification algorithm for both XR
and CT images to perform automated diagnosis of COVID-19 cases. The obtained
results in this thesis reveal that the combination between discrete wavelet and discrete
curvelet transforms gives the best noise reduction results with most medical image
modalities except with Us images. The segmentation results reveal that the proposed
FCM with morphological operations and active contour segmentation achieve the
best segmentation results with approximately all images. The classification results of
the statistical approach revealed success of the statistical approach especially with
Us and CT images. The simulation results prove that the CNN model gives the best
classification results due to the ability of the CNN to extract a group of features based
on different convolution masks.