Classifying Brain Tumor in MRIs Using Spatial Attention-Based Adaptive Inception Residual Net and Localizing Abnormalities via Hyperdense Inception 3D Trans-UNet + + Model

A brain tumor is described as the expansion of unusual brain tissues, in which some of these tissues develop as cancerous. Detecting the tumors within the brain at the starting phase is essential for the victim’s life span and therapy plans. Moreover, exact segmentation and categorization of the tumor tissues are significant for proposing appropriate therapy strategies based on the progression of the tumor. Although Magnetic Resonance Imaging (MRI) approaches are used to examine tumors as well as advanced techniques based on artificial intelligence are employed to design adaptive networks for recognizing and distinguishing the tumors using MRI images which remains a major drawback. Additionally, the classical segmentation technique utilizes more resources and time for segmenting tumors from MRI images. To overcome these issues, a novel brain tumor segmentation and classification model based on the concept of deep learning for treating the affected brain cancer cells is developed. Initially, the MRI images of the brain are gathered from internet resources. The collected images are segmented using the Hyperdense Inception 3D Trans-UNet[Formula: see text] (HI-3DT-UNet[Formula: see text] model. This model is effective in extracting multi-scale data and related details by exploiting the hyperdense connections and the factorization of 3D convolutional layers. Also, this network permits effective reutilization of features that enhances the correctness of segmentation and execution. Then, the segmented images are given to the proposed Spatial Attention-based Adaptive Inception Residual Net (SA-AIResNet) for classifying brain tumors. The spatial attention enables the network to focus on the most relevant regions in the image, whereas the adaptive inception model acquires a broad range of characteristics that lead to enhanced accuracy in classification. Moreover, the competence of the classification process is enhanced by optimizing hyperparameters within the SA-AIR network with the help of the Improved Random Parameter-aided Fossa Optimization Algorithm (IRP-FOA). It enhances the system performance and helps to eliminate overfitting issues, leading to quicker training and better exactness of the system. Thus, the designed system efficiently focuses on the needed area of segmented images by accurately classifying whether the brain tumor is malignant or normal by avoiding overfitting problems. The robustness of the introduced system is validated with other brain tumor detection models to validate its efficacy. The result shows that the proposed IRP-FOA-SA-AIResNet model achieves 95.50% accuracy, which is superior as compared to other traditional models and leads to better classification of brain tumors.