Classification of Brain Tumor Using Hybrid Pre-Trained Networks and MRMR Algorithm

Brain tumor classification is one of the crucial uses of medical image processing. A longer life is possible if the tumor is correctly and promptly diagnosed. Because manually segmenting brain tumors for cancer diagnosis is difficult and time-consuming, automatic classification of brain tumor images is necessary for this task. Pre-processing, feature extraction, feature selection, and classification are the four stages of the general framework for automatic tumor detection from MRI images. The method for automatic brain tumor classification described in this research combines machine learning and deep learning algorithms. We used three pre-trained deep networks to extract the most detailed information from MRI images. ResNet, AlexNet, and GoogleNet are the networks that were utilized for feature extraction for the diagnosis of brain tumors, the classification model has been a support vector machine (SVM). Additionally, in this research, the best feature vector was chosen using the MRMR algorithm to improve classification speed and accuracy. The BRATS database is used to provide the training dataset. The BRATS validation dataset showed promising results for the investigated method. This method's complete tumor classification accuracy on experimental data is 99.5% on average in this dataset.