Early Alzheimer’s Disease Detection by Using ResNet for Extraction of Features with SVM Tuning for Classification

Abstract

Early detection of Alzheimer's disease (AD) is vital for effective intervention and management. This study introduces a novel approach that employs a ResNet architecture for feature extraction from neuroimaging data, coupled with Support Vector Machine (SVM) tuning for classification. We utilized a dataset of MRI scans from subjects at various stages of cognitive decline. The ResNet model was fine-tuned to identify complex patterns in brain structure indicative of early AD. Features extracted from the network's final layers were then input into an SVM classifier, optimized through grid search to improve classification accuracy. The parameters are tuned and the final layer in Resnet-53 is used for Support Vector Machine (SVM) model. It is mainly used for detection of Alzheimer's, and it addresses the both binary and multiclass classification. But we are using multiclass labels are used so it is used for multiclass classifications.  Bayesian optimization with Hyperopt is used in a fine tune process to find the hyperparametrs space, optimizing key variables such as kernel selection and regularization to increase the performance of the model on the validation set. The proposed method is used to show the most accurate difference between normal cognitive aging and Alzheimer's disease, which is so sensitive and specific to achieve.  Multimodal data is combined to increase the models performance and provides comprehensive tools for early detection.