Scalable Synthetic Data Generation for HDLSS SERRS cTnI Detection via Conditional GMMs

Sahil Sharma; Thomas Poudevigne; Sayantan Tripathy; Ng Ka Wai; Muskaan Singh; LJ McDaid; Gerard L. Coté; Samuel B. Mabbott; Saugat Bhattacharyya

doi:10.1109/SENSORS59705.2025.11330562

Scalable Synthetic Data Generation for HDLSS SERRS cTnI Detection via Conditional GMMs

Sahil Sharma
, Thomas Poudevigne
, Sayantan Tripathy
, Ng Ka Wai
, Muskaan Singh
, LJ McDaid
, Gerard L. Coté
, Samuel B. Mabbott
, Saugat Bhattacharyya

Texas A and M University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

18 Downloads (Pure)

Abstract

The diagnosis of Myocardial Infarction (MI) can be enhanced through label-free biosensing techniques like Surface-Enhanced Resonance Raman Spectroscopy (SERRS). Machine learning is hindered by data scarcity, excessive dimensionality,
and class imbalance. This paper presents a scalable and interpretable pipeline for generating synthetic data for high-dimensional, low-sample-size (HDLSS) SERRS data. The method combines class-conditional Gaussian Mixture Models (GMMs), physiological renoising through Savitzky–Golay filtering and Gaussian noise, and manifold-constrained filtering using UMAP-KNN to maintain physiological validity. A composite score that integrates fidelity (Wasserstein), diversity (DTW), and generalisability (UMAP acceptance) is used to rank synthetic data configurations during generation. Experimental results indicate that classifiers trained over 5x-scaled synthetic data attain F1 scores of up to 82.35% and ROC AUC of 1.00, surpassing the performance of models trained solely on real data. This lightweight, modular approach offers a scalable alternative to
deep generative models in the context of clinical biosensing.

Original language	English
Title of host publication	IEEE SENSORS Conference 2025
Publisher	IEEE Xplore
Pages	1-4
Number of pages	4
ISBN (Electronic)	979-8-3315-4467-6
ISBN (Print)	979-8-3315-4468-3
DOIs	https://doi.org/10.1109/SENSORS59705.2025.11330562
Publication status	Published online - 19 Jan 2026
Event	2025 IEEE SENSORS - Vancouver, BC, Canada Duration: 19 Oct 2025 → 22 Oct 2025

Publication series

Name
ISSN (Print)	1930-0395
ISSN (Electronic)	2168-9229

Conference

Conference	2025 IEEE SENSORS
Country/Territory	Canada
City	Vancouver, BC
Period	19/10/25 → 22/10/25

Bibliographical note

Publisher Copyright:
© 2025 IEEE.

Funding

This work has been supported by the Department for the Economy Northern Ireland under the US-Ireland R&D Partnership Programme (USI-207), and by the NSF’s Engineering Research Center for Precise Advanced Technologies and Health Systems for Underserved Populations (PATHS-UP), #1648451.

Funders	Funder number
Department for the Economy
US-IRELAND R&D Partnership Programme	USI-207
NSF	1648451

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 9 Industry, Innovation, and Infrastructure
SDG 12 Responsible Consumption and Production

Keywords

Synthetic Data
SERRS
Gaussian mixture model
Cardiac Troponin I
High-dimensional Data
point-of-care (PoC) diagnostics
Gaussian Mixture Model
Point-of-Care Diagnostics
High-Dimensional Data

Access to Document

10.1109/SENSORS59705.2025.11330562

Accepted PaperAuthor Accepted Manuscript, 3.08 MB

Cite this

@inproceedings{1090889e32084f2089138274b373b1e9,

title = "Scalable Synthetic Data Generation for HDLSS SERRS cTnI Detection via Conditional GMMs",

abstract = "The diagnosis of Myocardial Infarction (MI) can be enhanced through label-free biosensing techniques like Surface-Enhanced Resonance Raman Spectroscopy (SERRS). Machine learning is hindered by data scarcity, excessive dimensionality, and class imbalance. This paper presents a scalable and interpretable pipeline for generating synthetic data for high-dimensional, low-sample-size (HDLSS) SERRS data. The method combines class-conditional Gaussian Mixture Models (GMMs), physiological renoising through Savitzky–Golay filtering and Gaussian noise, and manifold-constrained filtering using UMAP-KNN to maintain physiological validity. A composite score that integrates fidelity (Wasserstein), diversity (DTW), and generalisability (UMAP acceptance) is used to rank synthetic data configurations during generation. Experimental results indicate that classifiers trained over 5x-scaled synthetic data attain F1 scores of up to 82.35\% and ROC AUC of 1.00, surpassing the performance of models trained solely on real data. This lightweight, modular approach offers a scalable alternative to deep generative models in the context of clinical biosensing.",

keywords = "Synthetic Data, SERRS, Gaussian mixture model, Cardiac Troponin I, High-dimensional Data, point-of-care (PoC) diagnostics, Gaussian Mixture Model, Point-of-Care Diagnostics, High-Dimensional Data",

author = "Sahil Sharma and Thomas Poudevigne and Sayantan Tripathy and Wai, \{Ng Ka\} and Muskaan Singh and LJ McDaid and Cot{\'e}, \{Gerard L.\} and Mabbott, \{Samuel B.\} and Saugat Bhattacharyya",

note = "Publisher Copyright: {\textcopyright} 2025 IEEE.; 2025 IEEE SENSORS ; Conference date: 19-10-2025 Through 22-10-2025",

year = "2026",

month = jan,

day = "19",

doi = "10.1109/SENSORS59705.2025.11330562",

language = "English",

isbn = "979-8-3315-4468-3",

publisher = "IEEE Xplore",

pages = "1--4",

booktitle = "IEEE SENSORS Conference 2025",

address = "United States",

}

TY - GEN

T1 - Scalable Synthetic Data Generation for HDLSS SERRS cTnI Detection via Conditional GMMs

AU - Sharma, Sahil

AU - Poudevigne, Thomas

AU - Tripathy, Sayantan

AU - Wai, Ng Ka

AU - Singh, Muskaan

AU - McDaid, LJ

AU - Coté, Gerard L.

AU - Mabbott, Samuel B.

AU - Bhattacharyya, Saugat

PY - 2026/1/19

Y1 - 2026/1/19

N2 - The diagnosis of Myocardial Infarction (MI) can be enhanced through label-free biosensing techniques like Surface-Enhanced Resonance Raman Spectroscopy (SERRS). Machine learning is hindered by data scarcity, excessive dimensionality, and class imbalance. This paper presents a scalable and interpretable pipeline for generating synthetic data for high-dimensional, low-sample-size (HDLSS) SERRS data. The method combines class-conditional Gaussian Mixture Models (GMMs), physiological renoising through Savitzky–Golay filtering and Gaussian noise, and manifold-constrained filtering using UMAP-KNN to maintain physiological validity. A composite score that integrates fidelity (Wasserstein), diversity (DTW), and generalisability (UMAP acceptance) is used to rank synthetic data configurations during generation. Experimental results indicate that classifiers trained over 5x-scaled synthetic data attain F1 scores of up to 82.35% and ROC AUC of 1.00, surpassing the performance of models trained solely on real data. This lightweight, modular approach offers a scalable alternative to deep generative models in the context of clinical biosensing.

AB - The diagnosis of Myocardial Infarction (MI) can be enhanced through label-free biosensing techniques like Surface-Enhanced Resonance Raman Spectroscopy (SERRS). Machine learning is hindered by data scarcity, excessive dimensionality, and class imbalance. This paper presents a scalable and interpretable pipeline for generating synthetic data for high-dimensional, low-sample-size (HDLSS) SERRS data. The method combines class-conditional Gaussian Mixture Models (GMMs), physiological renoising through Savitzky–Golay filtering and Gaussian noise, and manifold-constrained filtering using UMAP-KNN to maintain physiological validity. A composite score that integrates fidelity (Wasserstein), diversity (DTW), and generalisability (UMAP acceptance) is used to rank synthetic data configurations during generation. Experimental results indicate that classifiers trained over 5x-scaled synthetic data attain F1 scores of up to 82.35% and ROC AUC of 1.00, surpassing the performance of models trained solely on real data. This lightweight, modular approach offers a scalable alternative to deep generative models in the context of clinical biosensing.

KW - Synthetic Data

KW - SERRS

KW - Gaussian mixture model

KW - Cardiac Troponin I

KW - High-dimensional Data

KW - point-of-care (PoC) diagnostics

KW - Gaussian Mixture Model

KW - Point-of-Care Diagnostics

KW - High-Dimensional Data

UR - https://pure.ulster.ac.uk/en/publications/1090889e-3208-4f20-8913-8274b373b1e9

UR - https://www.scopus.com/pages/publications/105034134667

U2 - 10.1109/SENSORS59705.2025.11330562

DO - 10.1109/SENSORS59705.2025.11330562

M3 - Conference contribution

SN - 979-8-3315-4468-3

SP - 1

EP - 4

BT - IEEE SENSORS Conference 2025

PB - IEEE Xplore

T2 - 2025 IEEE SENSORS

Y2 - 19 October 2025 through 22 October 2025

ER -

Scalable Synthetic Data Generation for HDLSS SERRS cTnI Detection via Conditional GMMs

Abstract

Publication series

Conference

Bibliographical note

Funding

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this