الفهرس | Only 14 pages are availabe for public view |
Abstract Early detection and accurate staging of a cancer is considered an important issue in practical radiology. It can increase the possibility of human survive. Medical imaging is the process of human body visualization. It plays an essential role in human body diagnosis and treatment. Much information can be obtained from dierent medical imaging techniques like CT and MRI, which it needs an expert to evaluate and analysis these types of images in short time. Also, it is quite common that for same set of medical imaging, dierent doctors may come up with dierent diagnosis results. Computer-Aided Diagnosis (CAD) has become one of the major research subjects in diagnostic medical imaging. The basic concept of CAD is to provide a fully automatic system to assist radiologists in medical image interpretation. Moreover, it can help surgeons through identifying the location and size of tumors. The main objective of the thesis is to introduce a modied computer-aided diagnosis (CAD) system in medical imaging, since medical images are relevant sources of information for detecting and diagnosing a large number of illnesses and abnormalities. Due to their importance, this thesis is focused on several medical imaging type including (1) CT liver images, (2) chromosomes histopathology slide imaging, (3) breast histopathology slide imaging and (4) breast thermal imaging. The proposed CAD system is mainly divided into ve sub CAD systems. Each sub CAD system is dealing with dierent medical imaging modalities such as CT, Thermal and Histopathology. Moreover it is divided into four parts: (1) organ region segmentation, (2) detection and extraction of abnormalities regions, (3) features extraction for abnormalities candidates regions and nally (4) candidate classication to normal or abnormal (benign or malignant). All the proposed sub CAD systems in this thesis are mainly depend on two approaches; Neutrosophic Sets and Swarm Optimization approaches. In the rst investigation in this thesis, a fully automatic mitosis detection and counting system for breast cancer histopathology slide imaging using Neutrosophic Sets has been presented. In this system the candidates have been extracted using Neutrosophic image. In order to enhance the extracted candidate, morphological operators have been used which helps in eliminating too small regions that are non-mitosis. Several features have been extracted from the detected candidates that are focused on statistical, texture and shape features. The second study presented a system for automatic classication of thermogram imaging to normal or abnormal. This system consists of two main phases: (1) automatic segmentation done by Neutrosophic sets in conjunction with fuzzy c-means to get region v of interest (ROI); (2) classication achieved by extracting several features, i.e. statistical, texture and energy, and then classied by SVM to into normal and abnormal. Interphase cells are undivided and the condensedmass of chromosomes. They can highly decrease the eciency of automatic karyotype. Therefore, a new fully automatic system based on fast fuzzy c-means (FFCM) and grey wolf swarms optimization (GWO) has been presented. The proposed system used to remove interphase cells and extract chromosomes from metaphase chromosomes image. It comprised of three phases, namely; preprocessing, chromosomes image clustering and post-processing phase. The obtained results show a good performance of the proposed system. The fourth study presented a fully automatic CAD system for CT liver image diagnosis based on using GWO. The proposed system comprised of four phases: (1) liver segmentation, (2) lesion segmentation, (3) features extraction from each candidate and nally (4) candidates lesion classication. The experimental results show the eciency of the proposed system. Moreover, it has been compared with other systems. Finally, an integrated system based on using PSO approach and watershed algorithm for automatic liver segmentation from abdominal CT images has been proposed. Several measurements are used to evaluate the performance of the proposed systems for dierent applications. These measurements are Correlation, Specicity, Dice Coecient, Jacard Index, Accuracy, Sensitivity, F-measure, True Positive Ratio, Error Rate, Precision and Recall. The experimental results show that the proposed liver segmentation system using PSO for 43 images obtains 94% and using GWO for 62 images obtains 96% and obtains 97% for liver diagnosis. The proposed detection and counting system for breast cancer histopathology slide using 35 benchmark histopathological data base images taken from ve dierent aspects obtains overall using Neutrosophic approach 77.61% F-score, Recall 74.92% and Precision 81.25 % with accuracy 90.1%. The proposed CAD system for breast cancer detection from thermal image using 61 breast cancer thermal bench mark dataset images using Neutrosophic approach obtains overall 92.06% Accuracy, 96.55% Recall, 87.50% Precision and 7.94% Error rate. The proposed segmentation and interphase cells removal system from chromosomes slide images GWO for 30 chromosomes slide images obtains overall 93.61% Accuracy, 90.99% Precision, 89.61% Sensitivity, 6.39% Misclassication Rate and 96.35% Specicity. |