An Intelligent Framework for Alzheimer's disease Classification Using EfficientNet Transfer Learning Model

Alzheimer's disease (AD) is a prevalent psychological disorder. The economic cost of treating for AD patients is expected to increase. Therefore in the last few years, research on AD diagnostic has laid great emphasis on computer-aided methods. The significance of developing an artificial intelligent diagnostic technique towards accurate and early AD classification seems essential. Deep-learning models hold significant benefits over machine learning approaches as these techniques do not require any kind of feature engineering. Moreover, T1-weighted Magnetic Resonance Imaging (MRI) is the neuroimaging data modality which is widely practiced for such a purpose. In some cases, the most significant barrier to integrating DL models into pre-existing applications is a lack of adequate data architecture. Changing medical information is usually hard to communicate, examine, and interpret. Transfer learning (TL) allows designers to use a combination of models in order to fine-tune a specified solution to a target problem. Transferring knowledge across two separate models could lead a generally a more reliable and precise model. In this work, researchers utilized an EfficientNet TL model already trained on ImageNet dataset to categorise subjects as AD vs. Cognitive Normal (CN) based on MRI scans of the brain. The dataset for this study was acquired from Alzheimer Disease Neuroimaging Initiative (ADNI). The performance parameters such as accuracy, AUC were used to evaluate the model. The proposed model on ADNI dataset achieved an accuracy level of 91.36% and AUC as 83% in comparison to other existing transfer learning models.