Quantitative Research

Ongoing research

SCOS Cuffless Blood-Pressure Research

Timothy constructed a project-specific dataset from raw scientific-trial data, engineered more than 700 cardiac-waveform features, and adapted an existing XGBoost workflow under Ariane Garrett's advisement and with advisor-assisted code.

Multi-panel scientific figure showing SCOS optical measurement principles, cardiac waveforms, and simultaneous measurement configurations.
Outcome
Established an analysis-ready dataset and feature-engineering workflow for ongoing cuffless blood-pressure research, with preliminary findings presented internally.
Timothy’s contribution
Timothy built the project-specific dataset, engineered more than 700 waveform features, performed clinical-cohort analysis, contributed blood-pressure prediction and hypertension-classification work, and developed CAD for a wearable test base.
Role
Undergraduate Researcher
Institution
Boston University Biomedical Optical Technologies Lab
Date
February 2026–Present
Technologies
Python, XGBoost, Autodesk Inventor
Methods
Dataset construction from raw trial data, Cardiac-waveform feature engineering, Blood-pressure prediction, Hypertension classification, Clinical-cohort comparison, CAD test-base development
Visit institutional contextRelated experience: Undergraduate Researcher

Advisor-authored publication informing the work

Ariane Garrett, Byungchan Kim, Nil Z. Gurel, Edbert J. Sie, Benjamin K. Wilson, Francesco Marsili, John P. Forman, Naomi M. Hamburg, David A. Boas, and Darren Roblyer, “Speckle contrast optical spectroscopy for cuffless blood pressure estimation based on microvascular blood flow and volume oscillations,” Biomedical Optics Express 16, 3004–3016 (2025).

View publication (opens in a new tab)

Problem and constraints

Cuffless blood-pressure research requires measurements collected through scientific trials to become structured, comparable inputs for analysis. Timothy's project sits in that translation layer: preparing a dataset for his research question, deriving informative features from cardiac waveforms, comparing clinical cohorts, and adapting an established modeling workflow without overstating what the current evidence supports.

The public case study therefore separates implemented capabilities from clinical conclusions. It does not claim clinically validated performance, publish participant-level data, or present unapproved prediction or classification results.

Timothy's role

As an undergraduate researcher, Timothy owns the project-specific dataset construction and feature-engineering work described here. He also contributes clinical analysis, model-workflow adaptation, cohort comparison, CAD, and internal technical presentation. Garrett provides postdoctoral advisement, trial context, and code support for the existing XGBoost workflow.

Collaboration

The workflow depends on close interpretation between advisor-led experimental work and Timothy's downstream analysis. Garrett's trial data and guidance establish the scientific context; Timothy converts that material into a project-specific analytical structure, develops features and comparisons, and brings preliminary findings back to the lab team for discussion.

Technical architecture and methods

Dataset construction

Timothy organized raw data produced through advisor-run scientific trials into a project-specific dataset suitable for repeated analysis. This step established the structure needed to compare clinical cohorts and connect measured cardiac waveforms with downstream blood-pressure and hypertension-related tasks.

Feature and modeling workflow

Timothy engineered more than 700 features from cardiac waveforms. He then adapted an existing XGBoost workflow to his dataset under Garrett's advisement and with advisor-assisted code. The workflow supports ongoing blood-pressure prediction and hypertension-classification investigations, but this case study intentionally reports no model-performance numbers.

Physical development

The research also includes a physical interface for testing. Timothy developed CAD in Autodesk Inventor for an attachable wearable test base, connecting the quantitative work to the practical requirements of repeatable measurement hardware.

Engineering decisions and research constraints

Timothy's contribution emphasizes traceable preparation and comparison rather than unsupported claims of model novelty. The existing XGBoost workflow is treated as advisor-supported infrastructure, while his dataset construction, feature engineering, cohort analysis, and project-specific adaptation are identified separately.

The same boundary governs public results. Preliminary findings have been presented to the lab team, but no clinical-validation claim or quantitative performance metric appears here while the research remains in progress.