Classification of alzheimer's and breast cancer images with deep learning methods

With early detection and diagnosis of the disease, the risk of death of patients is reduced. However, the disease can often be misdiagnosed during manual radiological examinations. For this reason, computer-aided methods that make fast and accurate decisions are needed. In this study, Alzheimer's and breast cancer diseases, which have an important place in recent years, were discussed and a study was developed for the early diagnosis of these diseases. Within the scope of the thesis, deep learning models were applied on the sample data set to detect Alzheimer's disease and the results were compared. In the first stage of the study, the sample data set was analyzed and it was determined which features would affect Alzheimer's disease. In the next stage, a multi-class model was created by detecting Alzheimer's disease according to its stages (Mild_Demented, Moderate_Demented, Non_Demented, Very_Mild_Demented) with four different deep learning models. To classify breast cancer types (adenocarcinoma, large. cell. carcinoma, normal, squamous.cell.carcinoma), the dataset was normalized by going through various preprocessing processes. In the next stage, breast cancer was automatically detected according to the stages of the cancerous cell. In addition, the performance evaluation of different deep learning models such as DNN, CNN, VGG16 and AlexNet for the diagnosis of Aizhemer and breast cancer was carried out. In addition, the network structure created for all models was analyzed with different distributions (Glorot Uniform, HeNormal). Results were obtained with different epoch values. In the first stage, all proposed models were run for 100 epochs and results were obtained. In the second stage, all models were evaluated by taking the results for the 6 epochs with the highest success rate according to the "val_loss" criterion. This study, unlike other studies, increased the performance and success rates for both data sets by developing the common network structure of deep learning architectures. In addition, the classification success of the proposed models, modeled with different network structures, for both data was analyzed with different evaluation criteria (precision, accuracy, sensitivity, F1 score) and the results were evaluated, and the complexity matrices were analyzed for training, validation and testing for all models. All models were evaluated and solution suggestions were presented to facilitate and guide scientists working in this field in disease prediction and detection. Keywords: Accuracy, Alzheimer, Breast Cancer, Classification, Deep Learning,