الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 131 |  |  |
| | | from | 131 | | |
Abstract
Breast cancer (BC) is a common health problem of major significance, as it is the most widespread kind of cancer among women which leads to morbidity and mortality. Pathological diagnosis is considered as the golden standard of BC detection. However, the investigation of histopathology images is a challenging task. Automatic diagnosis of BC could lower the death rate by constructing a computer-aided diagnosis (CAD) system capable of accurately diagnosing BC and reducing the time consumed by pathologists during examinations. In deep learning, this can be often by and huge done by extricating features through a convolutional neural network (CNN) and at the moment classifying employing a very associated organizer. This Thesis presents two different CAD systems to classify BC to benign and malignant. The first proposed CAD method consists of 4 stages; image pre-processing, feature extraction and fusion, feature reduction, and classification. The first CAD is based on fusion features extracted with ResNet Deep Convolution Neural Network (DCNN) with features of wavelets packet decomposition (WPD) and histograms of oriented gradient (HOG). Next, the feature data were reduced by utilizing principal component analysis (PCA). Finally, the reduced features are used to train different individual classifiers. Results show that the highest accuracy of 97.1% is achieved. In the second proposed CAD method, 3 different feature fusion strategies are used. In the first strategy, all deep features ((ResNet, AlexNet, GoogleNet) and WPD are fused. In the second strategy, feature sets are ranked descendingly according to the classification accuracy achieved by the classification step. Afterward, a sequential forward strategy is employed to fuse feature sets and selects the combination of fused feature sets which improves the accuracy of the proposed CAD. In the third strategy, the sequential backward method is used to fuse feature sets and select the appropriate fused feature set that impacts the performance of the CAD.Then, a multiple classifier technique for each feature set is applied. Next, Deep Learning fusion trained by Multiple classifier fusion was applied. After that, the features is reordered to the highest accuracy and forward selection is applied, the probabilities between features is combined, fusion between features trained by Multiple classifier fusion results is applied. Finally, Deep Learning fusion with backward selection trained by Multiple classifier fusion Results was applied. This fusion is made to examine theb influence of feature fusion on the performance of the proposed CAD system. Results show that the highest accuracy of 98.86% is achieved.The results for the two proposed CAD systems were compared with recent related CAD systems. The comparison showed that our proposed CAD systems are capable of accurately classifying BC to benign and malignant compared to other work and have competing performance. Thus, they can be used helpfully in medical experiments and investigation procedures.