An enhanced deep learning approach for cancer diagnosis /
أسلوب مجود فى التعليم العميق لتشخيص السرطان
Sarah Ali Abdelaziz Ismael ; Supervised Hesham A. Hefny , Ammar Mohammed
- Cairo : Sarah Ali Abdelaziz Ismael , 2020
- 87 P . : charts , facsmilies ; 30cm
Thesis (M.Sc.) - Cairo University - Faculty of Graduate Studies for Statistical Research - Department of Computer and Information Science
Cancer is the second leading cause of death after cardiovascular diseases. Out of all types of cancer, brain cancer has the lowest survival rate. Proper diagnosis of the tumor type enables the doctor to make the correct treatment choice and help saving the patients life. Human diagnosis based on MRI images is an unreliable, error-prone, time-consuming, highly specialized task that depends on the experience of the radiologist. Misdiagnosis of brain tumor type can prevent the effective response to medical treatment and decrease the chance of survival among patients. Therefore, there is a high need in the Artificial Intelligence field for a highly accurate Computer Assisted Diagnosis (CAD) system to assist doctors and radiologists with the diagnosis and classification of tumors. Traditional machine learning approaches required feature extraction which is a manual and time-consuming task that requires prior knowledge about the problem domain. Over recent years, deep learning has shown an optimistic performance in computer vision systems in the medical imaging domain In this thesis, we propose an enhanced approach for classifying brain tumor types using Residual Networks. The proposed model is evaluated on a benchmark dataset containing 3064 MRI images of 3 brain tumor types. We have achieved the highest accuracy of 99% outperforming the other previous work. The results were evaluated under several performance metrics such as accuracy, precision, recall, f1-score, and balanced accuracy. Furthermore, the results have been evaluated on two different distributions of the dataset, an image-level approach, where images were split randomly into training and validation sets. And a patientlevel approach where a patient is present in either in the training set or validation set but not both