Model Proposal of DEMXNET for the Diagnosis of Alzheimer's Disease and Comparison with Deep Transfer Learning Methods

Purpose: Studies on AI-assisted MRI analysis for early detection of Alzheimer's disease (AD) have recently offered remarkable results. This study aims to classify Alzheimer's disease with state-of-the-art deep learning methods and the new DEMXNET model we propose, and comparative results are presented.MethodsWe used the synthetic Minority Oversampling Technique (SMOTE) because of the class imbalance problem in the data set. MRI images were fine-tuned to deep learning models VGG16, VGG19, InceptionV3, Xception, and DenseNet201, and our suggested model DEMXNET was used to classify Alzheimer's type dementia disease, and the performances of the models were compared.Results: Accuracy rates of VGG16 (%92.89), VGG19 (%95.94), InceptionV3 (%89.65), Xception (%91.05), DenseNet201 (%95.47), and our suggested model DEMXNET (%96.48) were obtained in the classification problem of Alzheimer's type dementia disease.Conclusions: Due to the class imbalance problem of the data set we used in the study, SMOTE was used to improve the accuracy of all models used, and the accuracy rate was increased in all models. The DEMXNET model we suggested was more successful than all models. For Imbalanced datasets, the proposed method can process efficiently and reduce the computational cost.