Research portfolio
24 projects across EHR data science, medical imaging, and radiation oncology treatment planning.
Longitudinal Trends in Guideline-Concordant Care After Peripheral Artery Disease Diagnosis
Tracked four cardiovascular disease prevention measures in 1.21 million patients with peripheral artery disease and examined how care changed over time and across patient groups.
Peripheral artery disease (PAD) detection model for deployment at University of California San Diego
Preparing an explainable peripheral artery disease detection model for clinical deployment using Epic data and Amazon Web Services SageMaker.
Performance, generalizability, and fairness of a peripheral artery disease detection model
Tested the model across five University of California Health systems, including performance by patient phenotype and demographic group.
Peripheral artery disease detection using structured EHR data
Developed the model in the American Family Cohort and tested it on independent UC San Diego Health data.
Diabetic foot ulcer and amputation risk: prediction and causal inference
Early-stage project combining diabetic foot complication risk prediction with causal analysis of foot exams and amputation outcomes.
Representation learning for longitudinal cardiovascular risk prediction
Learning patient representations from longitudinal EHR sequences for later prediction of stroke, amputation, myocardial infarction, and related events.
Variation in peripheral artery disease care across University of California Health systems
Compared patient populations, provider involvement, guideline-directed therapy, and limb outcomes across five health systems.
Registering prostate MRI to histopathology for imaging biomarker validation
Built a multimodal registration pipeline aligning in-vivo prostate MRI with whole-mount histopathology to create pathology-grounded biomarker labels.
Calibrating the RSI restriction score across echo times
Developed and validated a calibration method for the Restriction Spectrum Imaging restriction score in 197 prostate MRI cases.
Validating synthetic high b-value diffusion-weighted imaging for prostate cancer detection
Compared synthetic and acquired diffusion-weighted images in 151 prostate MRI cases, finding signal underestimation and pelvic artifacts that limit quantitative use.
ReIGNITE: Clinical validation of diffusion MRI-guided prostate tumor targeting
Prospective international reader study showing that quantitative restriction maps improved tumor contour accuracy, reduced complete misses, and made focal-boost targeting more consistent.
ReIGNITE: Clinical impact of prostate tumor contour variability
Converted multi-reader prostate tumor contours into focal boost plans to quantify how contour variability changes dose coverage and predicted biochemical-failure risk.
Validating MRI-derived synthetic computed tomography for radiotherapy planning
Compared synthetic and clinical planning images across 43 brain radiotherapy plans, including registration, dose recalculation, and failure-mode review.
Automated brachytherapy treatment planning
Built the dose-to-plan layer that converted predicted 3D dose into deliverable cervical cancer brachytherapy plans and tested it against clinical plans.
Knowledge-based OAR dose prediction for cervical brachytherapy
Co-developed and validated a MATLAB model that used patient anatomy to predict achievable bladder, rectum, and sigmoid doses for plan-quality assessment.
Knowledge-based brachytherapy dose prediction for ORBIT-RT
Translated a MATLAB model into a one-click MIM Software extension that displayed predicted bladder, rectum, and sigmoid doses from patient contours.
Predicting dose differences between cervical brachytherapy applicators
Applied separate tandem-and-ring and tandem-and-ovoid dose models across cohorts and checked the predicted organ-at-risk differences against clinical plans.
Predicting the need for interstitial needles in cervical brachytherapy
Developed and validated models that predicted whether OAR constraints could be met without needles, using standardized no-needle replans as the reference.
Automated ARIA batch export
Developed and deployed a Java application that retrieved patient data from the ARIA oncology information system using a supplied medical record number list.
EMT-integrated Flexitron system for breast brachytherapy verification
Upgraded the C# measurement software and built the MATLAB registration and reconstruction pipeline, validated in phantoms and 20 patient implants.
EMT-based patient study of catheter stability in breast brachytherapy
Designed the clinical workflow, performed 355 tracking measurements across 41 patients, analyzed the data, and presented the findings to clinical physicists.
Decision framework for selective adaptive planning in breast brachytherapy
Used Plastimatch for deformable registration and contour propagation, reconstructed catheters and reapplied plans in OncentraBrachy, and analyzed the results in MATLAB to design a physician-informed adaptive-replanning cascade.
Quantifying observer variability in brachytherapy catheter reconstruction
Designed a 13-patient, 426-catheter study and performed breast brachytherapy treatment planning to determine how reconstruction variability translates into dwell-position variations and dosimetric changes.
EMT-based automatic catheter reconstruction for brachytherapy planning
Built a C# workflow that registered EMT measurements to CT, exported catheter reconstructions as DICOM, and transferred them into OncentraBrachy.