Residual attention network for brain tumor classification using MRI images

The brain acts as the primary region of the human nervous system that helps to perform different functions in the human body. The irregular proliferation of brain cells, which can lead to death if left untreated, is termed a brain tumor. Nowadays, deep learning frameworks have been utilized to detect brain tumors from medical imaging patterns for increasing the patient prognosis and survival. Meanwhile, the prevailing techniques employed for detecting brain tumors encountered overfitting issues and also typically require large databases. Initially, the pre-processing of input MRI images is executed by image de-noising and image enhancement techniques. Then, the enhanced image is applied to Attention Deeplabv3+ (Att-Deeplabv3+) for segmenting brain tumor regions, and the extraction of various features is executed from the segmented brain tumor region. The derived feature set is then employed by the ResNA-Net architecture to perform the brain tumor detection task. The ResNA-Net attained superior performance in detecting brain tumors, and the ResNA-Net achieved specificity, sensitivity, accuracy, precision, Matthews Correlation Coefficient (MCC), and F1-score of 96.220%, 97.468%, 96.776%, 95.988%, 96.289%, and 96.722%. This study introduces a hybrid deep learning framework, termed the Residual Neuron Attention Network (ResNA-Net), for the early detection of brain tumors using MRI brain images. By integrating attention-based tumor segmentation with residual neural network-based classification, the proposed approach offers an effective solution for improving the accuracy and reliability of automated brain tumor detection.