Caste-Net: A Cross-Attention Swin Transformer With Dynamic Adaptive Algae Optimization For Explainable Multiclass Dementia Diagnosis From Brain Mri

Abstract

Dementia is a disorder of the brain's neurons and making a good diagnosis and prediction of the disorder early may yield better treatment that will prevent the loss. Disease detection and classification requires optimized solutions, based on deep learning techniques, are complex. Motivated by these challenges, this study presents a novel Cross-Attention Swin Transformer Efficient Network (CASTE-Net)-based framework. The model's hyper-parameters are optimized using the improved optimization algorithm, resulting in enhanced learning capability and superior classification performance. Furthermore, noises can be eliminated from the pre processing procedure of the MRI images via the combination of B-spline and Lanczos Interpolation without loss of the sensitive details.Image enhancement is first performed to enhance the quality of the MRI scans and enable robust feature extraction. The enhanced images are then decomposed into multi-scale frequency components to exploit both spatial and spectral information associated with brain abnormalities, after which attention is applied to extract deep features.To further reduce the complexity of the extracted features, and to improve the accuracy of the classification, the proposed model is adopted to select the optimal feature and, consequently, greatly reduces the processing overhead complexity in the selected features.The MRI images and their feature representation, along with the intricate patterns embedded in the MRI images, are used for the classification by the developed Cross-Attention Swin Transformer Efficient Network (CASTE-Net). The accuracy of the benchmark OASIS dementia dataset was 98% and various parameters were studied using the developed Quantum Convolutional Neural Network. With the developed work this guarantees a very accurate classification of dementia.