An Effective Performance Analysis of the Alzheimer Detection System Using The Deep Features Incorporated Swin Transformer Model

Alzheimer's is a neurological disease that leads to severe memory loss in patients. This article employs a novel systematic workflow, the Alzheimer Detection System (ADS), for Alzheimer and Non-Alzheimer brain MRI image classification process, which uses the proposed Deep Feature Incorporated Swin Transformer Model (DFISTM). This proposed DFISTM is designed with edge detection, pixel-transformation process, convolutional feature computation, and the Swin Transformer Model (STM) classifier. The edge pixels in the brain MRI images are detected through the Mamdani fuzzy logic, and they are fused using Mamdani fuzzy rules to enhance the pixel intensity. Then the curvelet transform is applied on the detected edge pixels in the image in order to obtain the decomposed curvelet coefficients, where the decomposition produced low- and high-frequency band components from the fused brain MRI. The proposed deep learning-based feature extraction module, Dynamic Feature Enhanced Convolutional Network (DFECN), is used to compute the deep convolutional features from the two-dimensional curvelet coefficient matrix. These features are then classified into either Alzheimer or non-Alzheimer brain MRI images using the proposed STM classification approach. The novelty of this research work is to construct a DFECN model to compute the fine-grained feature maps with a minimum number of internal convolutional layers. This research article uses two Alzheimer datasets, Mendeley and Kaggle to train, validate, and test the proposed system with respect to the parameters Alzheimer Detection Sensitivity (ADSe), Alzheimer Detection Specificity (ADSp), Alzheimer Detection Precision (ADP), and Jaccard Index (JI).