Avi Surit°

Houston, TX U.S. Permanent Resident Authorized to work in the U.S.

APPLIED HEALTHCARE RESEARCH / HADASSAH HEART INSTITUTE

Unsupervised Cluster Analysis for Outcomes Prediction in LVH

A collaborative applied research project using UK Biobank clinical and cardiac MRI data to identify distinct phenotypes among patients with increased left ventricular mass and evaluate their association with clinical outcomes.

4,255patients with increased LVM
57clinical and MRI-based features
4distinct clinical subgroups
5 yearsmedian follow-up

THE WORK

From heterogeneous patient data to clinically meaningful phenotypes

The team prepared and analyzed large-scale UK Biobank data, combining demographic, clinical, laboratory and imaging-derived variables. Hierarchical clustering with Ward's minimum variance method was used to identify patient subgroups, followed by survival analysis to assess outcome differences.

The project also explored image-derived representations from cardiac MRI using a convolutional autoencoder and compared image-based clusters with clusters derived from clinical data.

KEY FINDING

Four clinically distinct clusters showed progressively different mortality risk and outcome profiles.

The analysis demonstrated that phenotype-based subgrouping can reveal clinically relevant structure beyond conventional geometry-based LVH classification.

COLLABORATION

Research team

Conducted by Ariel Vishne, Avi Surit and Ranel Loutati, in collaboration with David Luria, Offer Amir and Yitschak Biton from the Hadassah Heart Institute, within an applied M.Sc. project at the Hebrew University of Jerusalem.