Classifying and Localizing Abnormalities in Brain MRI Using Channel Attention Based Semi-Bayesian Ensemble Voting Mechanism and Convolutional Auto-Encoder

Brain tumors represent a severe and often life-threatening condition in adults, as the rapid multiplication of cancerous cells within the tumor can critically impair a patient’s normal functioning. The clinical practice commonly utilizes imaging modalities such as MRI, PET and CT scans to assess brain tumor’s size, type, and location. The purpose of this research is to create a computer aided diagnosis (CAD) system that can segment and categorize brain tumors automatically. The system is designed to work specifically with T1W-CE Magnetic Resonance Images (MRI) of the brain. The classification task involves determining the type of tumor present in the image, while the segmentation task involves separating the tumor region from the surrounding healthy tissue. By automating these tasks, the proposed system aims to increase the accuracy and the effectiveness of brain tumor diagnosis and treatment planning for patients. The multi-class classification of brain tumors (BCT) is considered one of the most daunting problems in medical imaging. This research article proposes a model named VS-BEAM that can be used efficiently for clinical decision-making. The proposed VS-BEAM (Voting Based Semi-Supervised Bayesian Ensemble Attention Mechanism) model has been examined for a brain tumor’s multi-class classification. The VS-BEAM model achieved the highest level of accuracy possible. The proposed work achieves maximum sensitivity, specificity, and diagnostic accuracy compared to existing models using T1W-CE MRI images. A convolutional autoencoder is utilized for extracting tumors from MRI images. The accuracy obtained from testing data of 264 brain tumors was 98.91%, indicating that the method is effective and can be used in the clinical context to assist in detecting larger or even smaller tumors.