ABSTRACT
The detection of the brain tumors accurately is crucial in terms of the diagnosis of the disease, treatment planning and evaluation of the results. Although surgery is the most widely used treatment method for brain tumors, the use of magnetic resonance scans with a non-invasive approach plays a significant role in the success of clinical practices. In this study, full automatic detection of brain tumors on magnetic resonance (MR) images in the prepared data set is proposed using faster R-CNN. Within the scope of the study, the effect of image pre-processing on detection performance is also investigated with experimental studies both without any pre-processing and with pre-processing by removing noise using a second-order Volterra filter and median filter (MF). In the experimental studies performed on MR images in the test set, 85.87% accuracy is obtained in the detection of brain tumors without any pre-processing. On the other hand, 88.70% and 91.06% accuracy rates are achieved in the detection of brain tumors with pre-processing by MF and second-order Volterra filter, respectively. Thus, it is determined that the performance of the proposed faster R-CNN-based method in the detection of the tumor region is increased since the tumor regions in the image pre-processed brain MR images are more apparent.
