Adaptive mobilenetv3 with spatial attention-aided effective brain tumor classification approach using denoised image-based segmentation

Abstract

The collected images are processed through an image denoising module, where the Sparse Autoencoder (SAE) is employed for denoising images. The denoised images are then used in the brain tumor segmentation module, utilizing Cascaded MobileUnet++ (CMUnet++) for accurate localization of tumor regions. The segmented images are subsequently fed into the Adaptive MobileNetV3 with Spatial Attention (AMSA) for classification performance. Parameters from MobileNetV3 are optimized using Modified Random Integer-based Fossa Optimization (MRI-FO) to design an AMSA model for brain tumor classification with high accuracy. The inclusion of Spatial Attention (SA) helps the model prioritize relevant features in the segmented images. This model maintains stable performance even when tumors are subtle or difficult to detect. The accuracy of the MRI-FO-AMSA approach exceeds that of 72% CNN, 89% SVM, 86% VGG-16, and 86% AMSA. The proposed technique enhances brain tumor detection by effectively analyzing complex and irregular tumor shapes with high accuracy. This approach has the potential to improve brain tumor diagnosis accuracy, leading to better patient outcomes through timely and precise treatment.