A robust classification of brain tumor disease in MRI using twin-attention based dense convolutional auto-encoder

Brain cancer is a life-threatening disease that affects many people and is caused by an abnormal growth of tissue in or around the brain. Therefore, early diagnosis and treatment of brain tumor are necessary. Detecting medical diseases is one of the crucial tasks in the clinical field as it helps improve the lives of patients. Magnetic resonance imaging (MRI) is generally used to diagnose brain tumors. However, the existing MRI-based deep learning (DL) model is a time-consuming process and produces less accurate results. To solve this problem, the proposed study uses a twin attention-based dense convolutional autoencoder (TA-CAE) to identify brain tumors in the MRI images. The input image is first captured from the dataset and then pre-processed through enhanced average filtering to remove unwanted noise and improve image quality through image resizing and HSV color channel conversion. The pre-processed images are then segmented to find the affected region. Here, segmentation is performed using Gannet-based Kapurs Thresholding (G-KaT) techniques. Oriented gradient pyramidal histograms and grayscale run length matrix (PHOG-GLRLM) feature extraction techniques were used to extract the shape and texture features of the MRI after segmentation. Depending on these features, the novel TA-CAE model diagnosed the brain tumor and classified it into three different brain tumor types such as glioma, pituitary tumor and meningioma. The Python tool is used for the simulation process, and the TA-CAE model is evaluated based on several performance metrics. The simulated results demonstrate that the proposed TA-CAE provides an accuracy of 97.28%, which is a better performance compared to other existing brain tumor classification techniques.