Table 2.

Example Applications of Machine Learning for Cognitive Phenotyping in People with HIV

Study Goals	Dataset	Data Type	Methods	Reference
Identify similarities in cognitive profiles among PWH and determine features associated with cognitive profiles	Cognitive testing and associated sociodemographic and clinical data	Multisite clinical and cognitive testing (N = 1646)	Self-organizing maps and clustering (unsupervised deep learning); random forest (supervised)	Dastgheyb et al [15]
Determine predictors of cognitive impairment subtypes in PWH using sociodemographic, clinical, and cognitive test data	Cognitive testing and associated sociodemographic and clinical data	Multisite clinical and cognitive testing dataset (N = 370)	Univariate and multiple logistic regression, random forest models (supervised)	Tu et al [16]
Predict neurocognitive trajectories in children with perinatal HIV using demographics, clinical blood markers, and mental health indices	Cognitive testing and associated sociodemographic and clinical data	Multisite clinical and cognitive testing dataset (N = 285)	Gradient boosted multivariate regression; feature selection with SciKit and PDPBox (supervised)	Paul et al [17]
Determine relationships between sleep health and cognitive function based on HIV serostatus and investigate interpretation based on analytical approaches	Cognitive testing, questionnaires, actigraphy data	Multisite actigraphy, pulse oximetry, and cognitive testing dataset (N = 463)	Partial least-squares regression, multidimensional construct, and random forest (supervised); latent class analysis (unsupervised)	De Francesco et al. [18]
Classify HIV infection based on structural MRI data and associated regional volumetric data and determine which regions are implicated in HIV infection	MRI and associated diagnoses	Single-site MRI dataset (N = 310)	Multiple kernel learning; chained and single-step regularization and support vector machine (supervised)	Adeli et al [19]
Predict diagnosis and cognitive measures in individuals with alcohol use disorder and HIV using structural MRI	MRI and associated diagnoses	Single-site MRI dataset (N = 549)	Customized sparse logistic regression with joint feature-sample selection compared with joint feature-sample selection with sparse feature selection and support vector machine (supervised)	Adeli et al [20]
Predict HIV-associated cognitive impairment using clinical and MRI-derived features including grey matter volumes and white matter integrity	MRI, clinical features, and associated diagnoses	Single-site MRI and cognitive testing merged datasets (N = 101)	Support vector machine; feature selection with LASSO regression (supervised)	Xu et al [21]
Classify HIV and cognitive impairment status using minimally processed structural MRI	T1-weighted MRIs and associated diagnoses	Merged MRI datasets (N = 1449)	Convolutional neural network with domain-specific predictors (supervised deep learning)	Zhang and Zhao et al [22]
Determine resting state networks that differentiate between groups based on HIV serostatus and cognitive status	MRI and associated diagnoses	Merged MRI datasets (N = 1806)	Relief feature selection and convolutional neural network (supervised deep learning)	Luckett et al [23]
Classify frail status based on neuroimaging features (volumetric data, arterial spin labeling, resting state functional MRI)	MRI and associated diagnoses	Single-site MRI dataset (N = 105)	Gradient-boosted multivariate regression; feature selection with SciKit and PDPBox (supervised)	Paul et al [24]

Abbreviations: HIV, human immunodeficiency virus; MRI, magnetic resonance imaging; PWH, people with HIV.