الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 125 |  |  |
| | | from | 125 | | |
Abstract
Breast cancer is a major cause of death for women worldwide, and early detection through imaging is crucial. Histopathology is one imaging technique that produces images of body tissue diseases, and a dataset of histopathological benign and malignant images is used for prediction. Computer-aided diagnosis systems and convolutional neural network (CNN) techniques are used for early detection of breast tumors. This study aims to identify the most accurate CNN model for classifying benign and malignant tumors and compare the results obtained using different types of CNN models, as well as the impact of image augmentation on the results. The study is divided into two parts. The first part involves using different models on the 400× and 100× BreaKHis datasets. When using the 400× BreaKHis dataset, it was discovered that the DenseNet 121 model achieved the highest accuracy of 98%. While using the 100× BreaKHis dataset, ResNet 34 had the best accurate level of 99.6%. Consequently, it was found that increasing the size of the dataset before classification is not a good decision. The second part of the study aims to enhance the accuracy of tumor classification by improving the image quality of mammogram images. These enhanced images are then trained using the most accurate model identified in the first part which is ”ResNet 34”. The research explores various techniques such as spatial and frequency domain contrasting, smoothing, and sharpening with CNN algorithms on mammogram images. All of the classification spatial and frequency domain results were enhanced by the augmentation techniques we presented. The highest spatial domain enhancement algorithms achieved a classification accuracy rate of 97.7%. While, the best frequency domain improvement approach attained a 100% accuracy rate.