Research

Algorithms, visualizations, and the evidence behind them.

I build tools to extract measurements from cardiovascular MRI, design views that make complex flow interpretable, and evaluate what machine-learning outputs let us measure.

Explore the methods
4D Flow MRI in aortic dissection. My patient streamline visualization shows the measured velocity field across the cardiac cycle. Color indicates speed (0–130 cm/s).

Original methods for aortic dissection.

01 · Submitted to Radiology: Cardiothoracic Imaging

Detecting hemodynamically active tears

Algorithm development · Internal clinical evaluation

An interpretable 4D Flow MRI pipeline that projects velocity relative to lumen geometry to identify candidate communications across the dissection flap and estimate their size.

Explore the tear detector
Tear-detector workflow from segmented lumen geometry to a selected flow-active candidate

Machine learning and computational modeling

Image quality is only part of the question.

First-author conference work · SMRA 2025

Conditioned diffusion for medical images

A pilot study of MRI-conditioned diffusion for cross-modality image synthesis, using a modified MONAI diffusion model and paired-image registration.

The model was trained on 38 patients and tested on two. Registration failures and variability in generated intensities highlighted the importance of evaluating image statistics alongside visual quality. The small test set limits conclusions about anatomical and clinical reliability.

Conference citation

First-author conference work · USNCCM 2025

Evaluating AI-accelerated MRI

An evaluation of whether learned MRI reconstruction preserves downstream wall shear stress measurements, using paired processing and agreement analysis in 97 patients.

The study used retrospectively simulated acceleration. Prospective acquisition and regional validation remain next steps.

Read abstract (PDF)

Computational neuroscience · Pittsburgh, 2021

Variability in a neural network model

My undergraduate work with Chengcheng Huang examined variability suppression in a neural network model. I presented this work at MathFest and Pittsburgh’s Neuroscience Club Symposium. It was an early foundation for my interest in computational models and their relationship to biological signals.

Research history in my CV

Research interests

Toward useful, inspectable AI for science.

I want to explore how AI agents can help researchers check data quality, investigate conflicting measurements, and assemble evidence for a hypothesis. A useful system would preserve the connection between each conclusion and its underlying inputs.

One concrete starting point is to test whether an agent can flag inconsistent imaging outputs and direct a researcher to the relevant source evidence. That is a future research question, with evaluation against expert review still to be designed.

Publications and ongoing work