Multimodal MRI Radiomics and Machine Learning Identify Node-Level Structural and Functional Alterations in HIV-Associated Asymptomatic Neurocognitive Impairment.

Objective: In the era of combination antiretroviral therapy (cART), asymptomatic neurocognitive impairment (ANI) has become a common manifestation of HIV-associated neurocognitive disorders (HAND). Because ANI occurs without overt functional impairment, it may be overlooked in routine clinical practice despite being associated with an increased risk of subsequent symptomatic cognitive decline. This study aimed to develop a classification model for HIV-ANI by integrating node-level radiomic features derived from structural and functional MRI within a machine-learning framework, and to investigate the potential association between abnormalities in core cognitive network nodes and large-scale brain network dysfunction. Materials and Methods: A total of 67 individuals with HIV-associated asymptomatic neurocognitive impairment (HIV-ANI) and 70 cognitively intact individuals with HIV (HIV-IC) were included. Core regions of the default mode network (DMN), central executive network (CEN), and salience network (SN) were defined as regions of interest (ROIs). Node-level radiomic features were extracted from 3D T1-weighted imaging, amplitude of low-frequency fluctuation (ALFF), and regional homogeneity (ReHo) maps derived from resting-state fMRI. Unimodal, multimodal, and imaging-clinical fusion models were constructed using ElasticNet-regularized logistic regression after a stratified 7:3 training-test split. SHapley Additive exPlanations (SHAP) and a nomogram were used for model interpretation and individualized risk prediction. Results: The imaging-clinical fusion model achieved the highest AUC in the test set (0.735), with numerically better discriminative performance than the unimodal models. Functional features, particularly ReHo-derived features, contributed substantially to model prediction. SHAP analysis identified the medial prefrontal cortex, anterior and posterior cingulate cortices, and insula as key predictive regions. Clinical factors, including years of education and duration of HIV infection, also contributed to model performance. Conclusion: Structural and functional alterations are detectable within core cognitive network nodes at the HIV-ANI stage. The integration of multimodal MRI radiomic features with clinical factors showed potential for identifying HIV-ANI and provided preliminary evidence that node-level abnormalities may be associated with broader network dysfunction in early HAND. Further validation in larger, independent cohorts is warranted.