Abstract: Convolutional neural networks (CNN), which have been proven to be effective computational methods for identifying pictures, are of special interest in neuroscience research due to their significance in identifying Alzheimer's disease (AD). The most prevalent form of dementia in the elderly community is AD. The necessity of swift pathology of AD identification by magnetic resonance imaging (MRI) persists. Many pre-processing techniques have converted three alternative projections of T1-weighted volumetric MRI scans into 2D space. Four CNN models utilizing transfer learning, i.e., VGG-19, Wide ResNet 50-2, GoogleNet, and Inception v3, employ pre-processed MRI for generating the outcome values that the proposed ensemble model would later combine. The proposed prediction model employs an ensemble approach to produce fuzzy rankings of the fundamental classification approaches using the Gompertz function and automatically integrates the base models' decision results to arrive at ultimate forecasts on the test instances for the stage of AD, i.e., mild demented, moderate demented, non demented, and very mild demented. The system's reliability is demonstrated by the framework's outstanding results on publicly accessible MRI datasets. Achieving an accuracy of 99.22%, recall of 99.53, precision of 99.69, f1-score of 99.61, and AUC of 98, the suggested ensemble performs better than the other underlying base classifier. When combined with additional medical examinations, the proposed ensemble model will be a useful and effective diagnostic tool for MRI scans for AD.

Key words: Alzheimer’s disease, deep convolutional neural network, dementia diagnosis, ensemble learning, magnetic resonance imaging