Method of differential diagnosis of mild cognitive impairment of various origins: cross sectional study

Abstract

INTRODUCTION:The differential diagnosis of mild cognitive impairment (MCI), due to the high prevalence in the population and the rapid increase in incidence, is an urgent task. The most common causes leading to the development of cognitive impairment are Alzheimer’s disease (AD), cerebrovascular pathology, and their combination. AD usually manifests as an amnestic type of mild cognitive impairment (aMCI) at the pre-dementia stage. Subcortical vascular mild cognitive impairment (svMCI) is considered as the prodromal stage of subcortical vascular dementia. According to the results of pathomorphological studies, it was found that subfields of the hippocampal formation are selective vulnerability to AD, ischemia/hypoxia, and aging.Currently, using the FreeSurfer 6.0 software, it is possible to obtain quantitative indicators of the hippocampal formation subfieldsin vivo.The current trend in medicine is the development and implementation of new diagnostic solutions based on artificial intelligence and machine learning. One of the machine learning algorithms is binary logistic regression, which we used in the course of the study for the differential diagnosis of MCI of various origins.OBJECTIVE:To develop a method for the differential diagnosis of mil cognitive impairment of various origins.MATERIALS AND METHODS:The study included patients with the syndrome of mild cognitive impairment who were examined in the department of geriatric psychiatry of the V.M.Bekhterev National Medical Research Center for Psychiatry and Neurology, from which two groups were formed: group 1 — patients with aMCI, group 2 — patients with svMCI. Conditionally healthy volunteers, comparable in age, sex and level of education, made up the 3rdgroup (control). MRI examination was performed using a Excelart Vantage Atlas XGV magnetic resonance imaging system (Toshiba, Japan) with a magnetic field strength of 1.5 T, followed by MR morphometry of the subfields of the hippocampal formation.Statistics:Statistical analysis was carried out using data that was converted from a database in Microsoft Excel to the statistical package IBM SPSS 21. To develop a differential diagnosis method, based on the data obtained, the binary regression method and ROC analysis were used.RESULTS:Based on the obtained MR-morphometry data, a method was developed using the binary logistic regression equation. The value of p≥0.5 makes it possible to classify the patient to the aMCI group, and the value of p<0.5 — to the svMCI. The sensitivity of the method is 90%, the specificity is 80%, and the accuracy is 85%.DISCUSSION:Using binary logistic regression, the selection of variants of sets of variables (quantitative indicators) was carried out, for which ROC curves were constructed. The selection criterion was the area under the ROC curve — the AUC criterion (Area Under the Curve). The largest area under the curve (AUC=0.824) in the differential diagnosis of aMCI from svMCI was determined for the combination of the volume of the left subiculum and the thickness of the right entorhinal cortex.Since patients in the aMCI group have a significantly lower number of vascular foci than in the svMCI group (p<0.05), at the next stage, one more variable, the volume fraction, was added to the selected combination of two variables (volume of the left subiculum and thickness of the right entorhinal cortex) hypointense foci. When conducting an ROC analysis with a combination of three variables, an increase in AUC to 0.892 was noted. Further, using a combination of three variables and a binary logistic regression equation, a method for differential diagnosis of aMCI from svMCI was developed.CONCLUSION:The method of differential diagnosis based on binary logistic regression using MR morphometry data allows to distinguish patients with aMCI from patients with svMCI with high sensitivity and specificity.