Machine learning × medicine
Clinical AI,
grounded in the
whole patient.
I’m Ryan King, a Research Data Scientist at Houston Methodist’s Digital Health Institute. I build models that learn from health records, clinical language, and medical signals to better understand patient risk.
01 / Research directions
Connecting the data.
Understanding the patient.
My work asks how we can turn the information collected during routine care into useful, reliable models of health.
Longitudinal EHR · Survival modeling
Risk that evolves with the patient
Learning from changing clinical measurements to study cardiovascular disease and heart failure risk over time.
ECG · Clinical text · Medical imaging
Complementary views of health
Aligning signals, images, and language to learn clinical representations that transfer across tasks.
External validation · Clinical evaluation
Models that earn their place in care
Studying transportability, clinician–AI agreement, and reproducible evaluation beyond a single dataset.
The question behind the work
How do we make
clinical AI useful?
A good prediction is a starting point. I’m interested in whether a model works in a new health system, captures changes over time, and offers information clinicians can interpret.
At Houston Methodist, I work with Sadeer Al-Kindi and Khurram Nasir on these questions, bringing machine learning methods into close conversation with cardiovascular medicine.
More about my background02 / Selected publications
Methods, benchmarks,
and shared tools.
-
Conference
ML-FLYP: Improving EHR Model Transportability with Multi-Label Contrastive Fine-Tuning
Multi-label contrastive fine-tuning for transferring EHR models across healthcare settings.
-
Paper
An Efficient Contrastive Unimodal Pretraining Method for EHR Time Series Data
Efficient self-supervised representation learning for long clinical time series, with external evaluation on eICU.
-
Paper
MEDS Decentralized, Extensible Validation (MEDS-DEV) Benchmark: Establishing Reproducibility and Comparability in ML for Health
A community effort to make clinical ML evaluation reproducible and comparable across datasets.
-
Paper
Multimodal Pretraining of Medical Time Series and Notes
Aligning clinical measurements and notes through contrastive and masked prediction objectives for learning with limited labels.
03 / Open source
Research others can build on.
Tools and community contributions that make clinical machine learning easier to reproduce.
torchmimic
A PyTorch toolkit for machine learning with MIMIC electronic health record data.
Contributormeds-torch
Deep learning pipelines for experiments with the Medical Event Data Standard.
ContributorMEDS-Transforms
Composable preprocessing tools for standardized, reproducible health data workflows.
04 / Background
A path from service
to clinical science.
Before my Ph.D. in Computer Science at Texas A&M, I served as a U.S. Army Infantry Officer. That experience still shapes how I think about leadership, collaboration, and work with practical consequences.
The longer story- 2026–present
- Houston MethodistResearch Data Scientist · Digital Health Institute
- 2021–2026
- Texas A&M UniversityPh.D., Computer Science · Defended December 2025
- 2015–2019
- United States ArmyInfantry Officer · 1st Armored Division
Let’s connect
Good questions
make better science.
I welcome conversations about multimodal learning, cardiovascular AI, and translating clinical models into useful tools.