COMPUTER-AIDED DETECTION OF BRAIN TUMORS USING IMAGE PROCESSING TECHNIQUES

Abstract

Brain tumors are masses formed by the uncontrolled proliferation of cells in the brain. Brain tumors can be malignant or benign and can be fatal if not accurately identified at an early stage. Computer vision processing is used for early diagnosis, monitoring treatment response, and tumor classification. This study aims to detect brain tumors, a significant disease of our time, using image processing techniques. Preprocessing and data augmentation techniques were applied to a dataset of 253 images. Initially, CNNs were used for tumor detection, but transfer learning was employed for better results. Pre-trained VGG-16, DenseNet-121, ResNet-50, and MobileNet_V2 architectures were used. The model, adapted with transfer learning, achieved better performance with less data by adding a customized output layer for brain tumor detection. Experiments showed the best results with VGG-16, achieving 84.61% accuracy before data augmentation and 92.31% after augmentation. Compared to other studies, the post-augmentation accuracy rate was observed to be better than many others. The study also compares results from other deep learning architectures. Summarizing the current technological advancements in various tumor categories may help researchers understand future trends.