Performance Comparison Between Different Per-trained Models with Resnet-53 Using MRI and PET Scan Alzheimer’s Disease Image Dataset

In order to provide immediate support and medical care to identify Alzheimer's disease (AD) as early as possible. By analysing patterns and features in large datasets, these approaches can identify subtle changes in brain structure, function, or biomarkers that may indicate the presence of the Disease at an early stage. Early detection allows for timely intervention and treatment, potentially improving patient outcomes. Using MRI and PET scans image datasets particular to Alzheimer's disease. This study compares the performance of several pre-trained models, like VGG-16, VGG-19, RESNET 50, INCEPTION V3 and Desnet121 with the proposed model ResNet-53. The main goal is to assess and compare how well these models are able to discriminate between healthy people and people with AD. By comparing each model's accuracy and precision, we use transfer learning to optimize them all. The performance of the RESNET-53 is strong to classify the AD and the accuracy is 99.65%. Our findings showed significant differences in performance, with certain models exhibiting higher accuracy in particular imaging modalities. In the proposed model the preprocessing will be initialized by a zero centering process then combined Gaussian filter with bilateral filter. For feature extraction, ResNet is used for its residual connections. In the ResNet architecture first layers are freezed and the last 3 layers are customized for feature extraction. The study emphasizes how integrating deep learning approaches with a variety of imaging modalities may enhance diagnosis accuracy. The accuracy obtained using VGG 16, VGG 19, ResNet 101, RESNET 50, DenseNet 121 and Inception V3 models are 89.61%, 92.81%, 96.32%, 95.27%, 97.80% and 96.44%. The proposed model provides a classification accuracy of 99.65%. The proposed model ResNet 53 has more accuracy. “ResNet-53 outperforms baseline models, achieving a precision of 98.96%, recall of 95%, and an F1-score of 96.97%, which demonstrates its ability to handle class imbalance more effectively than previous approaches.