Advanced MRI in the Co-occurrence of Liver Disease and HIV Infection

Figure 1: Schematic of prior guided learning proposed to improve CEST imaging at a 3T MRI. DL, deep learning; CS: compressed sensing.

Liver-related concerns persist among people living with HIV (PLWH), despite antiretroviral therapy benefits. Non-alcoholic fatty liver disease (NAFLD) affects up to 50% of those with HIV, influenced by the virus and treatment. This study examines the coexistence of NAFLD and HIV, emphasizing their potential impact on brain health. The hypothesis suggests that individuals with both HIV and NAFLD are at a higher risk of persistent blood-brain barrier dysfunction, metabolic dysfunction, and microcirculation alterations, resulting in more severe brain injury and impaired cognitive function, especially in older individuals.

To test this hypothesis, we will employ non-invasive state-of-the-art MRI techniques in the brain. Specifically, we will develop a method with compress sensing (CS) for the readout part of clinical chemical exchange saturation (CEST) MRI and prior-based deep learning with a minimal training dataset to significantly accelerate CEST acquisition, enhancing signal-to-noise ratio (SNR), and reconstructing CEST exchange parameter maps with high fidelity. The study will compare neuroimaging metrics in PLWH with NAFLD versus those without, examining associations with blood biomarkers. Additionally, the impact of NAFLD on cognitive performance will be assessed, mediated by brain injury measured by CEST and other imaging metrics both cross-sectionally and longitudinally. We will further use a deep learning model with multimodal MRI and blood markers to differentiate HIV and NAFLD.

The significance of this research lies in its use of rapid and high-quality CEST imaging, providing increased sensitivity to disease pathology. By exploring associations between imaging metrics, blood markers, and cognitive scores, the study identifies gaps in knowledge of brain injury and disease progression, with potential implications for understanding other brain-related disorders. Additionally, the use of deep learning models may assist in differentiating brain-related abnormalities of NAFLD from HIV, given their significant clinical and imaging overlap.