Advancements in Brain Tumor Detection using Transfer Learning Model with CNN-based Multi-Head Attention for Tumor Type Classification

Brain Tumour (BT) is a common and aggressive disorder that leads to a very short life expectancy in high-grade grades. Therefore, treatment planning is the primary phase in enhancing the quality of patient care. Numerous image models like Computed Tomography (CT), ultrasound imaging, and Magnetic Resonance Imaging (MRI) were employed to assess a brain's cancer region. Compared to other image techniques, MRI images are used to analyze the tumour in the brain. Conversely, the enormous amount of data generated by MRI scans prevents the manual classification of non-tumours and tumours in a precise manner. However, it is time-consuming and requires knowledge to examine the MRI imaging. Currently, the development of Computer-Aided Diagnosis (CAD), Machine Learning (ML), and Deep Learning (DL) models permits the expert to recognize BT. This manuscript presents a Hybrid Artificial Intelligence-Based Brain Tumor Classification with Transfer Learning and Metaheuristic Optimization (HAIBTC-TLMO) model using MRI imaging. The HAIBTC-TLMO model aims to develop an effective and accurate method for classifying BTs. Image pre-processing begins with noise removal using a bilateral filter (BF), followed by skull removal through Otsu thresholding and morphological operations. Moreover, the proposed HAIBTC-TLMO model utilizes the NASNetMobile method for feature extraction to detect the BT region from the input image data. The hybrid Graph Convolutional Gated Recurrent Network (GCGRN) method is also employed for classification. Finally, the Spotted Hyena Optimizer (SHO) optimally adjusts the hyperparameters of the GCGRN method, resulting in improved classification performance. The experimental analysis of the HAIBTC-TLMO approach is conducted on a BT MRI dataset. The performance validation of the HAIBTC-TLMO approach demonstrated a superior accuracy output of 94.64% over existing methods.