Deep Learning Analysis of Cerebral Blood Flow to Identify Cognitive Impairment and Frailty in Persons Living With HIV

Abstract

Background: Deep learning algorithms of cerebral blood flow were used to classify cognitive impairment and frailty in people living with HIV (PLWH). Feature extraction techniques identified brain regions that were the strongest predictors. Setting: Virologically suppressed (<50 copies/mL) PLWH (n = 125) on combination antiretroviral therapy were enrolled. Participants averaged 51.4 (11.4) years of age and 13.7 (2.8) years of education. Participants were administered a neuropsychological battery, assessed for frailty, and completed structural neuroimaging. Methods: Deep neural network (DNN) models were trained to classify PLWH as cognitively unimpaired or impaired based on neuropsychological tests (Hopkins Verbal Learning Test-Revised and Brief Visuospatial Memory Test-Revised, Trail making, Letter-Number Sequencing, Verbal Fluency, and Color Word Interference), as well as frail, prefrail, or nonfrail based on the Fried phenotype criteria (at least 3 of the following 5: weight loss, physical inactivity, exhaustion, grip strength, walking time). Results: DNNs classified individuals with cognitive impairment in the learning, memory, and executive domains with 82%–86% accuracy (0.81–0.87 AUC). Our model classified nonfrail, prefrail, and frail PLWH with 75% accuracy. The strongest predictors of cognitive impairment were cortical (parietal, occipital, and temporal) and subcortical (amygdala, caudate, and hippocampus) regions, whereas the strongest predictors of frailty were subcortical (amygdala, caudate, hippocampus, thalamus, pallidum, and cerebellum). Conclusions: DNN models achieved high accuracy in classifying cognitive impairment and frailty status in PLWH. Feature selection algorithms identified predictive regions in each domain and identified overlapping regions between cognitive impairment and frailty. Our results suggest frailty in HIV is primarily subcortical, whereas cognitive impairment in HIV involves subcortical and cortical brain regions.

Effects of Framingham 10-Year Cardiovascular Risk Score and Viral Load on Brain Integrity in Persons With HIV

“…Pseudocontinuous ASL (pCASL), a noninvasive technique that uses arterial blood water as a contrast medium, was used to measure CBF (TR = 3500 seconds, TE = 9.0 milliseconds, flip angle = 90°, labeling time = 1.5 seconds, postlabeling delay = 1.2 seconds, 90° flip angle, voxel size = 3.4 × 3.4 × 5.0 voxels, 64 × 64 acquisition matrix, 22 axial slices with a 1-mm gap). 21 Preprocessing involved registering each volume to a mean volume and calculating framewise displacement. Framewise displacement was computed from movement parameters extracted from the realignment, and pairs with more than 0.5-mm displacement between the control and label volumes were censored.…”

Modeling the Effects of HIV and Aging on Resting-State Networks Using Machine Learning

“…For all PLWH, neuropsychological testing targeted 5 neurocognitive domains previously described. 11 A total of 15 tests were administered including the following: (1) learning: total recall across the learning trials on the Hopkins Verbal Learning Test-Revised and Brief Visuospatial Memory Test-Revised; (2) memory: delayed recall on the Hopkins Verbal Learning Test-Revised and BVMT-R; (3) executive: Trail Making Test B, Letter–Number Sequencing, Verb Fluency, and Color Word Interference Test Trial 3; (4) motor/psychomotor speed: Trail Making Test A, Digit Symbol, grooved pegboard dominant and nondominant hands, and Symbol Search; and (5) language: letter fluency (FAS) and category fluency (animals). Time to completion or total correct served as the dependent measures in accord with standard methods.…”

“…Structural data preprocessing was completed using FreeSurfer, version 5.3 (http://surfer.nmr.mgh.harvard.edu) as previously described. 11 Visual inspection of automated segmentation results was performed for quality assurance, and corrections were made when necessary. RS-FC preprocessing methods were performed as previously described.…”

“…For all PLWH, neuropsychological testing targeted 5 neurocognitive domains previously described. 11 A total of 15 tests were administered including the following: (1) Normative cognitive status for the control participants was confirmed by a structured clinical rating scale (ie, the Clinical Dementia Rating scale 14 ) or online cognitive test protocol. 13 These tests results were only used to confirm normal cognition for the control participants.…”

“…Multiple studies have used ML to reveal novel mechanisms and associations relevant to HIV. 11,12 This study uses ML-based feature selection within a large cohort of PLWH on cART who were virologically controlled (,200 copies/mL) (n = 297) and 1509 controls to identify RSNs that distinguish between HIV serostatus (controls compared with cognitively normal PLWH (PLWH CN), and controls compared with cognitively impaired PLWH (PLWH CI) regarding different age bins (younger than 35 years, 35-55 years, and older than 55 years). Furthermore, we identified that the RSNs are the strongest predictors of cognitive status between PLWH CN and PLWH CI across different ages.…”