Competitive Multi-Class Alzheimer’s Disease Classification Using a Hybrid GAN–DWT–CNN Model on MRI Data

Abstract

To diagnose AD as it advances to a third pathological stage‚ MRI images of the brain still suffer from accuracy issues․ Subtle structure changes occur even at such an early stage of the disorder that they can barely be detected with existing technology․ In order to conduct multi-classification of Alzheimer disease cases from MRI images‚ this paper proposes an automatic diagnosis system of Alzheimer's based on Generative Adversarial Nets (GANs) and Discrete Wavelet Transform (DWT)․ Significantly as well as increasing the number of classes‚ has it been able to do this․ In addition to the data imbalance problem and the classification quality of deep learning algorithms‚ GAN-based augmentation can tackle changed scale of the features․ The DWT can also improve the discrimination power of multiresolution features extracted from MRI images classified by a deep learning model based on the four-class classification problem: Non-Demented‚ Very Mild Demented‚ Mild Demented‚ and Moderate Demented subjects․ The model achieved a precision of 92 percent for overall classification accuracy‚ and 91% when computing the macro-averaged F1-score across all classes․ What should be borne in mind is that the proposed framework has high recall on early stage patients (where other methods are difficult to apply)‚ which shows off its ability more․ The stability of the convergence of the training process of the model‚ together with the results of the confusion matrices‚ confirms the robustness of the model's use․ The stable process if training of the model together with analysis for the confusion matrices further confirms the stability of the proposed model and its training process․ The results of the experiments show that for diagnosing Alzheimer's disease‚ GAN-based augmentation and DWT feature extraction can provide a very important performance improvement‚ and that‚ overall‚ we can therefore believe that it is feasible to use the proposed framework as a computer-helped diagnosis tool in the clinical care of sufferers from this dread disease․