Computer-aided diagnosis of Alzheimer’s disease by MRI analysis and evolutionary computing

Purpose: Due to the potential impacts on public health and economy in general, Alzheimer’s disease, AD, has become one of the biggest challenges to science in the last decade. However, most studies that propose methods to support the diagnosis of the disease have difficulties that impact adoption in clinical practice. This work aims to present a system model to support the low-cost diagnosis of AD, aiming to facilitate adoption in clinical practice and with performance compatible with the best methods proposed in the state of the art. Method: Using MRI volumes from the ADNI-3 base, this study presents a model based on evolutionary computing and machine learning capable of identifying and using the most appropriate 2D slice set to maximize each of the metrics related to AD diagnosis. The MRI volumes are organized according to the following classes: Cognitively Normal (CN), Mild Cognitive Impairment (MCI) and Alzheimer’s disease (AD). Results: Through an optimal selection of slices, the best results for accuracy, sensitivity and specificity obtained in each class were, respectively: CN (90.97%, 94.25%, 91.93%), MCI (87.74%, 83.16%, 98.33%) and AD (94.52%, 93.94%, 96.39%) for female patients, and CN (89.35%, 93.10%, 88.79%), MCI (87.42%, 83.16%, 97.50%) and AD (93.87%, 90.91%, 96.03%) for male patients. Conclusion: This work presented a model to support the diagnosis of AD and MCI, showing good classification performance. The model uses image data extracted from a limited set of 2D MRI slices selected optimally by applying an evolutionary search algorithm. The model also uses demographic information from the population of patients with and without cognitive impairment. This work presented a way to use gender information to produce significant improvements in the diagnosis of Alzheimer’s disease. © 2021, Sociedade Brasileira de Engenharia Biomedica.