The effects of HIV infection and aging on brain structural and functional connectivity and neuropsychological performance (NP) remains poorly understood. Understanding the effects of the virus using a lifespan perspective is vital for providing appropriate care to people living with HIV (PLWH).
297 virologically well-controlled (<200 copies/mL) PLWH (mean age 48.5y, 67% male) on cART and 1509 HIV uninfected (HIV-) controls (mean age 45.4y, 62% male) matched for age, sex, and education were evaluated. All participants completed structural and functional neuroimaging. PLWH were classified as cognitively normal (HIVCN) or impaired (HIVCI) based on NP battery consisting of five domains. Relief feature selection identified the strongest predictive functional resting state networks (RSNs) with regards to HIV serostatus and degree of cognitive impairment within specific age bins (< 35, 35-55, and >55 years old). Deep learning models identified where the largest structural differences occurred in relation to the identified RSNs.
The Relief algorithm identified the strongest predictive RSNs of HIV status between HIV- controls and HIVCN as the salience (SAL) and parietal memory network (PMN). The strongest predictive RSNs of HIV status between HIV- controls and HIVCI were the SAL, PMN, and frontal parietal (FPN). The strongest predictive RSNs of HIV cognitive impairment status between HIVCN and HIVCI were the SAL, FPN, and ventral attention (VAN). When evaluating different age bins, FPN became a stronger predictor with age, SAL and VAN lost predictive strength, and PMN remained consistent. With structural neuroimaging, the primary differences in RSN topology occurred in cortical and subcortical regions, including the dorsal and rostral lateral prefrontal cortex (Figure 1), anterior cingulate, and caudate.
We have identified RSNs that discriminated HIV-, HIVCN, and HIVCI using machine learning. Our results suggest PMN and SAL are highly impacted by HIV, and additional involvement of FPN leads to increased cognitive impairment. Deep learning models identified where the largest differences in RSN topology occurred. When evaluating different age bins, variability in RSNs predictive strength was observed, even under viral suppression. These results suggest a complex set of RSN and structural changes that are unique to HIV status, aging, cognitive impairment status, and anatomy.
![]()