01

Measurement

Model how the observation is generated, including tissue mixing, sampling, preprocessing, technical nuisance, and modality-specific physics.

02

Representation

Recover latent state in a form that remains biologically interpretable enough to compare, test, and challenge.

03

Validation

Ask whether the representation survives held-out data, orthogonal measurements, alternative assumptions, negative controls, and identifiability checks.

Selected research

Current projects

Each project starts from a different measurement system, but the same standard applies: the computational representation should earn its biological interpretation.

Method development · MSK

HostBio / Comp-MRF

Can routine CT become a quantitative measurement of host tissue state?

I developed a scan-adaptive probabilistic framework for tissue composition that explicitly handles partial-volume mixing, acquisition-related nuisance components, and within-scan references. HostBio-derived thymic phenotypes have now been evaluated across 1,078 patients in four independent NSCLC cohorts, including randomized RTOG-0617, using immune, genomic, treatment, toxicity, response, and survival data.

1,078 patients · 4 cohorts · randomized trial data · MSK implementation

HostBio research story →
Thoracic CT views showing voxel-level posterior estimates of thymic tissue composition from HostBio Comp-MRF

M.S. thesis · Johns Hopkins

StageBridge

What can cross-sectional human tissue legitimately tell us about progression?

StageBridge represents regulatory state using transcription-factor and pathway activity, gene regulatory structure, and tissue context. A hidden recoverability failure forced a redesign from context-neutral to conditional transport, improving transition recovery from 0.27 to 0.84. The external PanIN analysis did not yield stable programs, although the limited number of donors prevented strong conclusions about calibration.

0.27 → 0.84 synthetic recoverability · donor holdout · external limitation

StageBridge research story →
StageBridge regulatory landscape over premalignant epithelial states
Model-supported regulatory-field geometry over observed AAH/AIS epithelial states. Click to explore StageBridge.

Published and conditionally accepted research · MSK

Secondary lymphedema

How do persistent inflammatory and epithelial programs contribute to chronic lymphatic disease?

My public work with the Mehrara laboratory includes a peer-reviewed study of keratinocyte and PAR2 biology in secondary lymphedema and a peer-reviewed review synthesizing systems-level and omics approaches to disease pathophysiology.

Clinical and Translational Medicine · Frontiers in Immunology · 2026

Published lymphedema work →
Published secondary lymphedema figure showing epidermal inflammatory remodeling and cell-specific PAR2 biology
Published secondary lymphedema research. Adapted from Park et al., Clinical and Translational Medicine (2026), CC BY 4.0.

Additional work

Methods and collaborations

Full CV →

AI and deep learning for thoracic imaging

Training and deployment support for nnU-Net v2 thymus segmentation and evaluation of pretrained Sybil lung-cancer risk predictions in 77 women, where predicted risk separated cancer cases from controls across all six prediction horizons.

Longitudinal imaging and dynamic prediction

Serial cone-beam CT, radiomics, tumor geometry, longitudinal-survival modeling, and dynamic prediction, including the MSK clinical-imaging arm of a collaborative mechanistic-modeling study of radiation-induced tumor migration and treatment scheduling.

Cancer genomics and reusable infrastructure

COSMIC-IMPACT and related workflows for MSK-IMPACT tumor sequencing, MSK-ACCESS longitudinal ctDNA, somatic alterations, clonality, mutational signatures, and reproducible multimodal analysis.