Publications

SfN 2024 Abstract #9832

Limbic and Whole‑Brain Functional Connectivity in Non‑Substance Abusers with HIV

Stuart D. Washington, Kehinde Omisore, Anisa Thomas, Cedric Cole, Tomilowo Abijo, Ashley VanMeter, and Marjorie Gondré‑Lewis

Human immunodeficiency virus (HIV) negatively impacts behavioral health and is co‑morbid with neurocognitive and psychiatric disorders, including substance use disorder (SUD). Neuroimaging studies repeatedly show diminished functional connectivity in people infected with HIV. However, previous studies appear to disregard any potential for HIV/SUD co‑morbidities, an oversight that represents a potential confound in HIV‑related neuroimaging literature. Further, the functional connectivity of limbic neural substrates underlying reward and SUD (e.g., nucleus accumbens, amygdala, and hippocampus) remain unexplored in people living with HIV (PLWH). Here, we obtained resting‑state functional magnetic resonance imaging (rsfMRI) data from a small population (N = 7) of PLWH who have no history of SUD. Functional connectivity in PLWH had generally reduced functional connectivity relative to healthy controls (N = 14), with the greatest differences occurring between visual cortex and cerebellum. Seed‑based analyses of left and right nucleus accumbens and hippocampus yielded robust connections with the default mode network in controls. Similar seed‑based analyses of the amygdala in controls yielded robust connections with inferior temporal lobe regions rather than the default mode network. Connectivity between corresponding regions in PLWH was reduced but recruited the default mode and inferior temporal networks. Our results suggest that (1) PLWH who do not have SUD show reduced overall functional connectivity relative to controls, consistent with previous rsfMRI studies of PLWH, and (2) this reduced connectivity in PLWH extends to limbic structures underlying reward, even in the absence of SUD.

SfN 2024 Abstract #11775

Understanding Addiction through AI: A Predictive Analysis Powered by Machine Learning Models

Dickson A. Acheampong, Stuart D. Washington, Dewayne A. Dixon, and Marjorie Gondré‑Lewis

Better predictive models are needed to identify people who are more likely to abuse substance or consume drugs such as opioids and cocaine because of the drug's prevalence and the linked socioeconomic and health impact effects that are linked to it. This study investigates how well the machine‑learning algorithm Lasso regression predicts substance abuse in people based on a variety of socioeconomic and demographic factors, such as age, sex, income, human immunodeficiency virus (HIV) status, and education level. We applied this model to a dataset of people classified as substance abusers with and without HIV and non‑users with and without HIV to help identify the most important predictors of substance abuse. The model’s levels of accuracy vary. By providing insights into the potential of machine‑learning techniques to identify high‑risk individuals and enable more targeted interventions, these results advance the study of substance‑use prediction. These data highlight how crucial it is to include cutting‑edge computational techniques into public‑health initiatives to address substance‑use disorders. Further, our work creates opportunities for further research to improve predictive models and investigate how well they work for various demographics.

Howard Research Day 2024 Abstract (Lecture)

Limbic and Whole‑Brain Functional Connectivity in Non‑Substance Abusers with HIV: 

Towards a Machine‑Learning Analysis of Neuroimaging Data

Stuart D. Washington

Human immunodeficiency virus (HIV) negatively impacts behavioral health and is co‑morbid with neurocognitive and psychiatric disorders, including substance use disorder (SUD). Neuroimaging studies repeatedly show diminished functional connectivity in people infected with HIV. However, previous studies appear to disregard any potential for HIV/SUD co‑morbidities, an oversight that represents a potential confound in HIV‑related neuroimaging literature. Further, the functional connectivity of limbic neural substrates underlying reward and SUD (e.g., nucleus accumbens, amygdala, and hippocampus) remain unexplored in people living with HIV (PLWH). Here, we obtained resting‑state functional magnetic resonance imaging (rsfMRI) data from a small population (N = 7) of PLWH who have no history of SUD. Functional connectivity in PLWH had generally reduced functional connectivity relative to healthy controls (N = 14), with the greatest differences occurring between visual cortex and cerebellum. Seed‑based analyses of left and right nucleus accumbens and hippocampus yielded robust connections with the default mode network in controls. Similar seed‑based analyses of the amygdala in controls yielded robust connections with inferior temporal lobe regions rather than the default mode network. Connectivity between corresponding regions in PLWH was reduced but recruited the default mode and inferior temporal networks. Machine‑learning algorithms further refined the neuroanatomical features of functional connectivity that distinguish controls from PLWH. Our results suggest that (1) PLWH who do not have SUD show reduced overall functional connectivity relative to controls, consistent with previous rsfMRI studies of PLWH, and (2) this reduced connectivity in PLWH extends to limbic structures underlying reward, even in the absence of SUD.

Howard Research Day 2024 Abstract (Poster)

Understanding Addiction through AI: A Predictive Analysis Powered by Machine Learning Models

Dickson A. Acheampong, Stuart D. Washington, and Marjorie Gondré‑Lewis

Better predictive models are needed to identify people who are more likely to abuse substance or consume drugs such as opioids and cocaine because of the drug's prevalence and the linked socioeconomic and health impact effects that are linked to it. This study investigates how well two machine‑learning algorithms—Lasso regression and Random Forest—predict substance abuse in people based on a variety of socioeconomic and demographic factors, such as age, sex, income, and education level. We applied these models to a dataset of people classified as substance abusers and non‑users to help identify the most important predictors of substance abuse. According to preliminary findings, the models' levels of accuracy vary. By providing insights into the potential of machine‑learning techniques to identify high‑risk individuals and enable more targeted interventions, this study advances the area of substance‑use prediction. The results highlight how crucial it is to include cutting‑edge computational techniques into public‑health initiatives to address substance‑use problems. Furthermore, our work creates opportunities for further research to improve predictive models and investigate how well they work for various demographics.