Brain Tumor Detection Based on Deep Learning Approach

Abstract

Detection of Brain Tumors is a challenging task in the medical field, as to evaluate the MRI images, experts in the field need to take time to pinpoint the issue of having a tumor. In the last few years, many artificial intelligence techniques have been employed to automate the process. The limitations of earlier methods were that they were trained on a single data set, so the system failed to detect brain tumors from diverse environments through MRI images. To overcome this issue and enhance efficiency, This research analyzes brain tumor detection using data augmentation and denoising techniques on medical images across three distinct data sets. Two DL Algorithms, EfficientNet, and ResNet50, were employed to evaluate the effectiveness of these techniques. These methods' efficiency was assessed using two DL Algorithms, EfficientNet and ResNet50. EfficientNet produced 100% training and 100% test accuracy on the first data set, small-bt-2c, whereas ResNet50 produced 100% training and 96% test accuracy. The second data set, large-bt-2c, yielded training accuracy with EfficientNet of 98.80%, test accuracy of 96.73%, and training accuracy with ResNet50 of 98.72% and test accuracy of 96.08%. The third dataset, large-bt-4c, produced training accuracy with EfficientNet of 99.93%, test accuracy of 96.32%, and training accuracy with ResNet50 of 99.44% and test accuracy of 95.09%. According to the findings, both models were quite good at spotting brain tumors, with EfficientNet often beating ResNet50. These results conclude that using data augmentation and denoising methods in DL Algorithms may significantly enhance the detection accuracy of brain tumors. By using cutting-edge machine learning methods, this study adds to continuing efforts in the medical community to improve the early and precise detection of brain tumors.