Research

Research Interests & Vision

As clinical AI models and agents rapidly advance within healthcare, bridging the gap between raw computational predictions and actual clinical workflows remains a central challenge in modern medicine. Building upon five years of continuous interdisciplinary research — transitioning from wet-lab experimental design to data-driven clinical AI development — my work focuses on designing reliable, simulated, and human-centered medical informatics systems that support clinicians in diagnosis and improve patient outcomes.

Research Summary

My work centers on developing multimodal neural network architectures, risk-aware clinical triage algorithms, and physics-informed computational models that bridge medical imaging, clinical informatics, and cellular biomechanics. The common thread is translating complex biomedical data and biological mechanisms into rigorous, interpretable computational tools and decision-support systems that enhance clinical efficiency and patient care.

Research Projects

DB-ATRG: Density and BI-RADS-Aware Triage and Automatic Report Generation for Mammography

Advisor: Dr. Russell Jeter • Affiliation: RIMMES Program

A human-centered clinical decision support framework (DB-ATRG) combining model-driven triage and automated report generation. Fine-tuned MedGemma 1.5 (4B) using QLoRA on DMID and VinDr-Mammo datasets, capturing 100% of high-risk malignancies within the first 20% of reading workload.

DB-ATRG QLoRA MedGemma 1.5 Clinical Decision Support SIAM 2026

Time Series Classification of Active Range of Motion in Robotics-Assisted Stroke Rehabilitation

Advisor: Dr. Russell Jeter • Affiliation: RIMMES Program

Supervised deep temporal learning and feature engineering framework (TCN, LSTM, 1D-CNN, InceptionTime, Transformers, TSFresh) analyzing 28+ million raw sensor kinematic windows from Motus Nova rehabilitation robotics to classify 10 active range of motion levels in stroke patients.

Kinematics Time-Series Motus Nova Robotics TCN & LSTM TSFresh Stroke Rehabilitation

Modeling Wound Healing Without Migration and Proliferation

Advisor: Dr. Yi Jiang • Affiliation: Center for the Advancement of Students & Alumni

A computational medicine framework simulating Retinal Pigment Epithelium (RPE) wound healing to explore therapeutics for Age-related Macular Degeneration (AMD). Simulated cellular purse-string contraction and cell fusion repair mechanisms in CompuCell3D via Q-Potts models.

CompuCell3D Q-Potts Model Monte Carlo AMD Therapeutics

Morphological Feature Analysis of Retinal Pigment Epithelial Cells During Aging

Advisor: Dr. Yi Jiang • Affiliation: Center for the Advancement of Students & Alumni

An automated biomedical imaging pipeline extracting 133 morphological and texture features from 326 RPE cell images to identify aging biomarkers, achieving a 90%+ cross-validation F1-score with a stacking ensemble (XGBoost, LightGBM, CatBoost).

Biomarker Discovery Feature Engineering Stacking Ensemble OpenCV

Enhanced PINNs for Collective Cancer Invasion

Advisor: Dr. Yi Jiang • Affiliation: Center for the Advancement of Students & Alumni

Engineered a Physics-Informed Neural Networks (PINNs) model to simulate collective cancer invasion, resulting in a data-efficient, mesh-free solution for complex Partial Differential Equations (PDEs) utilizing Tensorized Fourier Neural Operators (TFNO).

PINNs TFNO PyTorch Autograd Best Project Award (SURS 2025)

Laboratory Rearing Model of Lissachatina fulica for Acharan Sulfate Extraction & Evaluation

Topic: Natural Product Antimicrobial & Anticoagulant Assay

Established a laboratory rearing model of Lissachatina fulica using natural agents for 100% inhibition of pathogenic bacteria. Extracted acharan sulfate and demonstrated 89.40% anticoagulant activity and 32.09% thrombolytic activity.

Biomedical Sciences Gravimetric Analysis Clotting Assay Natural Products