Abstract
Modern pattern recognition and artificial intelligence systems can help in providing better health care and medical solutions. The performance of human diagnosis degrades due to fatigue, cognitive biases, systems faults, and distractions. However, artificial intelligence based diagnosis systems are less error prone and give safe support to clinicians in detection and decision making. This work presents a smart and reliable way of diagnosing Alzheimer's disease (AD) and its possible early stage i.e., mild cognitive impairment. Alzheimer's is a neurodegenerative disease and leads to severe memory loss and inability to cope with daily life tasks. The diagnosis of AD from structural images requires great skill and is challenging for human diagnostics. The presented framework is based on deep learning and detects Alzheimer's and its initial stages accurately from structural MRI scans. The framework analyzes four different classes simultaneously in a single setup. The testing accuracy of diagnosis obtained by the method is 98.88%. Experiments are also performed on binary data and transfer learning is applied for multiclass classification achieving 99.7% accuracy. Once a good trained model is obtained, the decision for an unseen test scan is given within a few seconds. These models are free from the factors that are responsible for causing human diagnostic errors. Therefore, these models are dependable and can provide much faster diagnosis.