Identification of mild cognitive impairment subtypes in the elderly using multimodality magnetic resonance imaging and unsupervised machine learning

• Grant number: 21/14873-2
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Effective date (Start): October 01, 2022
• Effective date (End): February 29, 2024
• Field of knowledge: Health Sciences - Medicine
• Principal Investigator: Carlos Ernesto Garrido Salmon
• Grantee: Rodolfo Dias Chiari Correia
• Host Institution: Faculdade de Filosofia, Ciências e Letras de Ribeirão Preto (FFCLRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil
• Associated scholarship(s): 23/00327-1 - Deep learning approach to predict conversion from mild cognitive impairment neuropsychological subtypes to Alzheimer's disease using MRI, BE.EP.PD

Abstract

The mild cognitive impairment (MCI) stage in the elderly is very comprehensive and complex, being also known to be a transitional stage between healthy aging and dementias such as, for example, Alzheimer's disease (AD). In 2011, a set of papers established some criteria for diagnosing what they called "mild cognitive impairment due to Alzheimer's disease". However, the researchers highlighted important limitations to the use of these criteria, such as the variability of these symptoms because they are mild or early and the absence of biomarkers. Consequently, one of the main effects of the uncertainties related to the diagnosis of CCL and early-stage dementias is low diagnostic accuracy. Given this scenario, some works have tried to evolve in better characterization of CCL by dividing it into subgroups, and with recent advances in artificial intelligence techniques, characterization involving multiple biomarkers and pattern identification has reached a new level. In this work, we intend to develop a computational tool to identify CCL subtypes and calculate for each subtype the probability to remain stable, to convert to AD or eventually to another type of dementia. For this, we first intend to use a set of attributes extracted from demographic, clinical and mainly MRI data of different modalities (structural MRI, diffusion-weighted MRI, functional MRI, etc.) to train, with unsupervised machine learning techniques, a model that best identifies subgroups. The aim is to find patterns in the data by grouping individuals that have common characteristics. Next, we will look for associations between the subgroups found and the results of neuropsychological tests. Finally, using longitudinal data, we will calculate the conversion rates for each subgroup. Thus, our approach may have great potential in the development of new quantitative biomarkers to aid in the accuracy of diagnosis, prognosis, and therapy. (AU)