UCSF

Pair Production Tomography (P2T) is an emerging imaging modality that detects pair production events from high-energy photon beams, providing unique atomic-number-based contrast. A fundamental challenge in P2T image reconstruction is positron range — the finite distance positrons travel before annihilating — which causes spatial blurring between the true pair production distribution and the reconstructed annihilation image. Unlike in Positron Emission Tomography (PET), the directed photon beam in P2T breaks the isotropy of positron emission, requiring a beam-direction-dependent correction approach. No positron range correction method has previously been developed for P2T.In this work, two positron range correction methods are developed and evaluated for P2T imaging: Richardson-Lucy deconvolution and a total variation (TV)-regularized optimization framework solved using the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA). Beamlet-specific positron range kernels were constructed from Geant4 Monte Carlo simulations using a 10MV bremsstrahlung source, capturing the joint distribution of transverse positron displacement and emission angle relative to the beam direction. Kernels were constructed at three resolutions — 2mm, 1mm, and 0.5mm — to evaluate the effect of kernel resolution on correction performance.Corrections were applied to all 46 beamlets of a simulated computed tomography (CT)-based phantom containing three high-Z inserts. Both methods consistently improved beam profile spatial accuracy across all 46 beamlets, reducing average beam full width at half maximum (FWHM) from 8.5mm to approximately 6mm. At the full image level, pseudo-high resolution reconstruction at 1mm followed by rebinning to the intrinsic 2mm detector resolution was found to be the optimal pipeline. Both Richardson-Lucy and FISTA reduced insert FWHM from 18mm to 16mm with the 1mm resolution kernel correction. Direct 2mm reconstruction and correction showed no improvement in insert FWHM, demonstrating the importance of kernel resolution for accurate correction.These results demonstrate that positron range correction is feasible for P2T imaging and can meaningfully improve spatial resolution. The current pipeline represents a practical and effective approach, providing a foundation for future development of spatially varying kernels and expansion to full-view phantom and patient data.

Plasma neurofilament light chain (NfL) is a sensitive blood-based biomarker of neuroaxonal injury with potential value for predicting Alzheimer’s disease (AD)-related neurodegeneration. However, plasma NfL is influenced by physiological factors, including age, body mass index (BMI), and renal function. Besides, it remains unclear whether adjusted NfL measurements can predict subsequent structural brain changes during the preclinical stage of AD. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), this study developed individualized plasma NfL reference distributions and evaluated their ability to predict longitudinal MRI-measured neurodegeneration. Generalized Additive Models for Location, Scale, and Shape were fitted in CU Aβ− reference participants with valid eGFR data (n=378) to derive age-, BMI-, and eGFR-adjusted NfL Z-scores. Linear mixed-effects models were then used to test whether baseline and longitudinal NfL Z-scores were associated with changes in hippocampal volume, AD-signature cortical thickness, and lateral ventricular volume among Aβ+ participants. Expected plasma NfL increased with age and lower eGFR and decreased with higher BMI. Adjusted NfL Z-scores increased progressively across the AD continuum. However, baseline NfL Z-score did not significantly predict longitudinal change in any MRI outcome in the CU Aβ+, cognitively impaired Aβ+, or Early AD Continuum cohorts. In an exploratory subset, a greater annualized increase in NfL Z-score was associated with faster hippocampal volume decline (β = −0.225, P = 0.045), although the association did not remain significant after multiple-comparison correction. These findings support physiological adjustment for individualized interpretation of plasma NfL. Although a single baseline measurement did not predict subsequent MRI change, longitudinal NfL dynamics may provide a more informative signal of ongoing neurodegeneration and warrant further evaluation in larger cohorts with more frequent repeated measurements.

Diffuse midline gliomas (DMGs) have infiltrative, poorly defined boundaries that make tumor measurement difficult and variable. Volumetric assessment may be more informative than bidimensional response measurements, but manual volumetry is time-consuming and automated methods typically provide a single volume without indicating reliability. We evaluated whether evidential deep learning (EDL) with model ensembling could generate calibrated 95% uncertainty intervals for automated DMG volumes.Two five-model EDL ensembles based on SegResNet and nnU-Net v2 were evaluated. Voxel-wise probability distributions were propagated into tumor-volume estimates and 95% intervals. Analyses included voxel-wise calibration, true-volume coverage, relative interval width, and comparison of SegResNet uncertainty with physician difficulty ratings and spatial uncertainty annotations.The EDL nnU-Net v2 Ensemble approached 95% coverage but produced intervals likely too broad for clinical use, whereas the EDL SegResNet Ensemble provided the strongest balance between coverage and precision. SegResNet intervals widened with physician-rated difficulty, while aleatoric and epistemic uncertainty localized near physician-identified uncertain regions and tumor boundaries. These findings suggest evidential ensembling can provide automated DMG volumetry with calibrated uncertainty estimates that reflect clinically recognizable sources of difficulty.

Congenital heart disease (CHD), a structural malformation of the heart or great vessels, is associated with brain abnormalities and an elevated risk of neurodevelopmental deficits. The mechanisms underlying these abnormalities remain incompletely characterized. The glymphatic system, the primary waste clearance system for the central nervous system, is driven by arterial pulsatility and dependent on systemic venous pressure. The glymphatic system may be impaired in CHD due to altered vascular pulsatility and systemic venous pressure resulting from abnormal cardiac physiology and surgical palliations. Diffusion tensor imaging analysis along the perivascular space (DTI-ALPS) is an MRI-based non-invasive metric that is used to evaluate glymphatic function and has not been previously applied to adolescents with CHD. This study includes 34 adolescents with critical CHD (Transposition of great arteries (TGA): n = 17, Single-ventricle physiology (SVP): n = 9, Other CHD: n = 8) and 99 healthy adolescent controls. DTI-ALPS indices were calculated using a publicly available diffusion MRI processing pipeline (alps.sh). The expected results of this study included lower DTI-ALPS indices in subjects with CHD compared to controls, with subjects with SVP expected to have significantly lower ALPS indices compared to those with TGA. Statistical analysis included Bland-Altman analysis, group comparisons using two-tailed t-tests, and multivariable linear regression adjusting for age. DTI-ALPS mean indices were significantly lower in the CHD group compared to controls (1.348 vs. 1.499; β = -0.152, 95% CI: -0.216 - -0.089, p < 0.001), and this difference remained significant after adjusting for age (β = -0.083, 95% CI: -0.118 - -0.049, p < 0.001). Additionally, each CHD sub-type demonstrated significant reduction in DTI-ALPS mean indices compared against healthy controls. No significant difference was found between TGA and SVP subgroups (p = 0.671). No significant relationship was found between DTI-ALPS index and age or sex within either CHD or control group. Overall, these findings represent preliminary evidence of potential glymphatic dysfunction in CHD patients, with DTI-ALPS being a novel non-invasive biomarker for brain health and potential future neuroprotective interventions. Further research with larger cohorts and neurodevelopmental outcomes is needed to determine the clinical implications of these findings.

The nucleus reticularis thalami (nRT) is a sheet of GABAergic neurons that surrounds the thalamus and plays critical roles in sensory processing, attention, and sleep. Despite its central position within thalamocortical circuits, the mechanisms by which GABAergic signals converge and integrate within the nRT remain incompletely understood. The goal of this dissertation was to investigate the organization and functional consequences of GABAergic signaling in the nRT, particularly that originating from the external globus pallidus (GPe), across circuit and behavior. In Chapter 2, we characterized a previously underappreciated projection from the GPe to the nRT. We found that GPe input exerts opposing effects across nRT subregions due to regional differences in chloride homeostasis, resulting in either inhibitory or excitatory responses to GPe stimulation. These opposing effects produce distinct downstream consequences in thalamocortical (TC) circuits with a potential role in motor control, thereby identifying the GPe→nRT pathway as a non-canonical route through which the basal ganglia influence thalamic information processing. In Chapter 3, we examined how nRT and TC neurons transform synaptic inputs into spiking outputs using synthetic current injections. We found that the same input currents elicited distinct firing responses in nRT and TC neurons, demonstrating that intrinsic cellular properties play a critical role in shaping firing output.Together, these studies demonstrate that the functional consequences of GABAergic signaling in thalamic circuits are shaped by multiple interacting mechanisms, including chloride homeostasis and intrinsic neuronal properties.

Understanding how white matter pathways in the brain, or the structural connectome (SC), shape and constrain functional dynamics is a central problem in neuroscience known as the structure-function problem. Using diffusion MRI and functional MRI from large publicly available datasets, including the Human Connectome Project, I developed a framework for studying the graph Laplacian of the SC and its relationship to brain function. First, I investigated the group-level properties of graph Laplacian harmonics, including which are shared across subjects versus unique to individuals, and their relationships to resting-state brain organization. I found that these harmonics form an integration-to-segregation spectrum, in which the ends of the spectrum are shared across subjects and localize to sensory regions, whereas the center is more individualized and relates to multimodal brain regions. Second, to better understand how observed brain function embeds an underlying notion of SC, I developed a graph deconvolution method that models and removes the effect of passive signal diffusion through the SC. I found that the deconvolved signals exhibit remodeled functional connectivity, decoupled resting-state networks, and a disruption of the typical unimodal-multimodal functional divide. Finally, to investigate whether passive diffusion can be used to recover information about the SC itself, I developed a model that parsimoniously parameterizes directed SC using gene gradients and fits this model to brain function. I found that this directed SC model can estimate ground-truth directed SC across three non-human species---C. elegans, mouse, and macaque---and that the resulting human directed SC estimates are biologically plausible and predict a directed net flow that reproduces the original unimodal-multimodal functional divide. Together, these findings position the graph Laplacian of the SC as a scaffold for brain function and offer a principled framework for studying brain function through the lens of biological structure.

Extremely premature infants exhibit substantial biological heterogeneity that contributes to differences in neonatal outcomes, including bronchopulmonary dysplasia (BPD), one of the most common complications among this population. However, the biological processes underlying this heterogeneity remain incompletely understood. While omics approaches provide powerful tools for investigating these processes, their application in extremely premature infants remains limited. This dissertation used metabolomic and genomic approaches to investigate the urinary metabolome of extremely premature infants at high risk of developing BPD by identifying metabolic dysregulation associated with BPD and determining how genetic variation captured by genomic ancestry proportions contributes to variation in metabolite levels. Metabolome-wide association analyses during the second and fourth weeks of life identified suggestive metabolite associations with BPD, leading to the enrichment of metabolic pathways involved primarily in amino acid and fatty acid metabolism. These patterns of metabolic dysregulation varied across time points and feeding modality (parenteral vs enteral nutrition), with most BPD-differential metabolites being at lower urinary levels during the fourth week of life in infants who develop BPD, particularly among those receiving parenteral nutrition. Beyond BPD-associated metabolic dysregulation, genomic analyses demonstrated that ancestry-linked genetic variation contributes to variation in the urinary metabolome of extremely premature infants. African genomic ancestry proportions were associated with histidine metabolites and enrichment of nucleotide metabolism pathway. Although limited overlap was observed between BPD-associated and ancestry-associated metabolic pathways, several metabolites showed nominal associations across both analyses, highlighting the potential contribution of ancestry-linked genetic variation to variability in clinically relevant metabolites. This dissertation demonstrates that the urinary metabolome of extremely premature infants is shaped by both BPD-associated metabolic dysregulation and ancestry-linked genetic variation, highlighting the value of integrating metabolomics and genomics to understand biological heterogeneity in this vulnerable population.

Motor control relies heavily on a functional network between the basal ganglia and the sensorimotor cortex. In Parkinson’s disease (PD), disruptions within this network can lead to hypokinetic motor symptoms, such as bradykinesia and rigidity. While deep brain stimulation (DBS) of the basal ganglia has been shown to mitigate these symptoms, the conventional constant-amplitude approach is limited in its therapeutic efficacy by its inability to adapt to fluctuating movement states and symptom severity. Sensing-enabled neurostimulators – which enable the detection of cortico-basal oscillations – have the potential to offer a dynamic alternative: closed-loop, movement-responsive DBS (mDBS) which delivers targeted stimulation during volitional motion. However, its translation to real-world environments requires machine learning (ML) models that can reliably decode movement across changing stimulation amplitudes and extended timeframes. To address this, we first evaluated the predictive power of cortico-basal signals during normal, daily activities under cDBS. Analyzing over 530 hours of continuous cortical and subcortical recordings from 15 PD patients (27 hemispheres), synchronized with wrist-worn accelerometer data, revealed distinct neural biomarkers: movement-related beta (13-30 Hz) desynchronization and gamma (40-80 Hz) synchronization in both regions consistently distinguished mobile from stationary states. Using these spectral features, we developed movement-predictive ML models. Cortical models outperformed subcortical ones, though combining both signals yielded the highest overall accuracy. Notably, while increasing stimulation amplitudes resulted in a decrease subcortical model performance, cortical and combined models remained resilient. These results support the feasibility of developing robust cortico-basal ML models for real-time movement decoding in naturalistic settings. Building on these findings, we conducted a blinded, randomized, crossover feasibility study in four PD patients evaluating chronic home use of subthalamic mDBS using embedded cortical ML models. The mDBS algorithm maintained precision and stability across different timescales (milliseconds to months) and settings (clinical and naturalistic). During structured motor tasks, mDBS increased forearm speed and prevented progressive bradykinetic slowing compared to cDBS, accompanied by cumulative changes in sensorimotor cortical beta activity and coherence. Additionally, during unsupervised, daily activities, mDBS significantly reduced bradykinesia severity compared to cDBS. The progressive symptom relief observed over hours of mDBS - and its subsequent reversal upon returning to cDBS - suggests that movement-responsive stimulation strengthens functional motor circuits and engages activity-dependent plasticity mechanisms rather than merely masking symptoms. Collectively, these findings support the feasibility and therapeutic efficacy of mDBS as a restorative neuromodulation strategy for movement disorders.

Violence epidemiology emerged as a federal public health science in the United States not from a laboratory or a policy agenda, but from the deaths of Black children in Atlanta, Georgia, between 1979 and 1981. Southern Roots: The Atlanta Child Murders and the Origin of Violence Epidemiology reconstructs how the federal response to the Atlanta missing and murdered children crisis produced the founding methods, categories, and institutional infrastructure of a new scientific field, and at what cost to the communities whose suffering generated its data.Drawing on previously unexamined archival collections at the Atlanta University Center’s Robert W. Woodruff Library Archives Research Center; Federal Bureau of Investigation’s (FBI) investigative records obtained through the Freedom of Information Act; Centers for Disease Control and Prevention (CDC) Epidemic Intelligence Service (EIS) Epidemiologic Assistance (Epi-Aid) Report EPI-81-16-2 and its unpublished tables; and original oral history interviews conducted by the author in Atlanta in February 2025 with the physicians who designed the public health investigation, this dissertation advances a single, sustained argument: the institutions commissioned with the role to protect children allowed their deaths were transformed into the data on which a new science was built. The dissertation proceeds in four chapters. Chapter One recovers and centers the community knowledge produced by Black Atlanta before federal authorities arrived, demonstrating that the structural analysis later claimed by public health was first articulated by bereaved parents and community members. Chapter 2 examines the ATKID case files (the bureau’s designation for the Atlanta Child Murders Investigation) as a living archive, showing how the FBI's thorough and successful federal investigation produced the data infrastructure that the CDC drew on directly when constructing its own epidemiological database. Chapter 3 explores the transformation of the murders into an epidemiological problem through the case-control study, showing how its methodological choices encoded structural conditions as behavioral risk factors. Chapter 4 follows the peer-reviewed literature that formalized those choices into the settled knowledge of a new subfield.The analysis is informed by Dorothy Porter's concept of health citizenship, Michel Foucault's theory of biopower, and Achille Mbembe's concept of necropolitics, while building on Samuel K. Roberts Jr.'s critical history of public health in Black communities in the US. Together these interpretive frameworks reveal how violence epidemiology emerged through the transformation of Black children’s deaths into epidemiological evidence and how a field dedicated to preventing violence was founded upon the structures of anti-Black racial inequality it failed to confront.

Tuberculosis (TB) remains one of the leading infectious causes of death worldwide despite the availability of curative multidrug therapy. Although standardized treatment regimens achieve high overall cure rates, substantial interindividual variability persists in drug exposure, treatment response, toxicity, and relapse risk. Furthermore, repeated failures of several treatment-shortening strategies in phase 3 clinical trials highlight persistent gaps in understanding the pharmacologic and biologic determinants of treatment outcomes. This dissertation investigated pharmacokinetic-pharmacodynamic (PKPD) and host genetic determinants of treatment response in drug-susceptible TB using data from the phase 3 Tuberculosis Trials Consortium (TBTC) Study 31/ACTG A5349 clinical trial.The first objective of this dissertation was to characterize exposure-response relationships for rifampicin, pyrazinamide, and moxifloxacin within contemporary multidrug treatment regimens. Population nonlinear mixed-effects models were developed to characterize drug pharmacokinetics and quantify interindividual variability in systemic exposure. Associations between drug exposure and clinical efficacy and safety outcomes were subsequently evaluated using regression and time-to-event modeling approaches.The rifampicin analyses demonstrated substantial variability in systemic exposure despite standardized dosing. However, rifampicin exposure within the observed range of the standard 600 mg daily dose was not associated with differences in stable culture conversion, tuberculosis-related unfavorable outcomes, or serious adverse events. Pharmacokinetic simulations further demonstrated that flat dosing produced exposure distributions comparable to current weight-banded dosing strategies, supporting the feasibility of simplified rifampicin dosing approaches within standard six-month therapy.The pyrazinamide analyses identified approximately seven-fold variability in systemic exposure among participants receiving standardized therapy. Both efficacy and safety outcomes were associated with pyrazinamide exposure, enabling identification of therapeutic exposure windows associated with favorable benefit-risk profiles. Flat dosing of pyrazinamide at 1000 mg daily was predicted to increase the proportion of participants achieving target exposures compared with current weight-banded dosing approaches. These findings provide clinically relevant evidence supporting future pyrazinamide dose optimization strategies.The moxifloxacin analyses evaluated whether variability in moxifloxacin exposure contributed to outcomes within the four-month rifapentine-moxifloxacin treatment-shortening regimen. Although substantial pharmacokinetic variability was observed and several demographic and clinical covariates influenced exposure, moxifloxacin exposure within the standard 400 mg daily dosing range was not associated with treatment efficacy or major safety outcomes. These findings suggest that variability in moxifloxacin exposure alone does not explain differential outcomes observed in fluoroquinolone-containing treatment-shortening regimens and reinforce the complexity of multidrug regimen performance in TB.The second objective of this dissertation was to investigate host genetic determinants of anti-TB pharmacokinetics and treatment response. Genome-wide and candidate pharmacogenetic analyses identified NAT2 acetylator status as the primary determinant of isoniazid clearance, while variants in AADAC and UGT1A1 influenced rifapentine and moxifloxacin pharmacokinetics, respectively. Importantly, NAT2 genotype differentiated treatment response across rifapentine-containing shortened regimens, with rapid and intermediate acetylators experiencing higher rates of tuberculosis-related unfavorable outcomes compared with slow acetylators. Addition of NAT2 genotype modestly improved prediction of unfavorable outcomes beyond baseline clinical characteristics alone. In contrast, no additional genetic variants demonstrated strong associations with efficacy or severe safety outcomes.Collectively, this dissertation demonstrates that substantial pharmacologic and biologic variability exists among patients receiving standardized tuberculosis therapy and that host pharmacogenetic factors may contribute to differential treatment responses, particularly in treatment-shortening regimens. At the same time, the findings illustrate that exposure-response relationships in multidrug TB therapy are highly complex and may not be fully explained by plasma drug exposure alone. By integrating PKPD and pharmacogenetic analyses within a large phase 3 clinical trial platform, this work advances understanding of the determinants of tuberculosis treatment response and supports future efforts toward individualized dosing strategies, biologically informed treatment optimization, and precision medicine approaches in tuberculosis therapeutics.