Classification of Brain Injury Severity Using a Hybrid broadband NIRS and DCS Instrument with a Machine Learning Approach - Proceedings of SPIE

Aug 5, 2026

Optical biomarkers of neonatal hypoxic ischemic (HI) brain injury enable continuous bedside assessment of injury severity, and existing research focuses on selecting physiological signals and feature combinations from optical measurements. This work introduces the FLORENCE multimodal platform combining hybrid broadband NIRS and diffusion correlation spectroscopy (DCS) with a machine-learning pipeline. In a preclinical newborn piglet model, the k-means based pipeline distinguishes control, mild, and severe HI groups, reporting 78% accuracy for controls vs HI, 90% for mild vs severe, and 80% for three-group classification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Abstract

Introduction

Background on neonatal hypoxic ischemic encephalopathy

Clinical context and hypothermia treatment

Optical biomarker research using NIRS and DCS

PROCEEDINGS OF SPIE [SPIEDigitalLibrary.org/conference-proceedings-of-spie](SPIEDigitalLibrary.org/conference-proceedings-of-spie) Classification of brain injury severity using a hybrid broadband NIRS and DCS instrument with a machine learning approach Danai Bili, Frédéric Lange, Kelly Harvey Jones, Veronika Parfentyeva, Turgut Durduran, et al. Danai Bili, Frédéric Lange, Kelly Harvey Jones, Veronika Parfentyeva, Turgut Durduran, Nikki Robertson, Subhabrata Mitra, Ilias Tachtsidis, "Classification of brain injury severity using a hybrid broadband NIRS and DCS instrument with a machine learning approach," Proc. SPIE 12628, Diffuse Optical Spectroscopy and Imaging IX, 126280D (9 August 2023); doi: 10.1117/12.2670657 Event: European Conferences on Biomedical Optics, 2023, Munich, Germany Downloaded From: [https://www.spiedigitallibrary.org/conference-proceedings-of-spie on](https://www.spiedigitallibrary.org/conference-proceedings-of-spie on) 19 Oct 2023 Terms of Use: [https://www.spiedigitallibrary.org/terms-of-use](https://www.spiedigitallibrary.org/terms-of-use) Classification of Brain Injury Severity Using a Hybrid broadband NIRS and DCS Instrument with a Machine Learning Approach Danai Bili*a, Frédéric Langea, Kelly Harvey Jonesb, Veronika Parfentyevac, Turgut Durdurand, Nikki Robertsonb, Subhabrata Mitrab, Ilias Tachtsidisa a Department of Medical Physics and Biomedical Engineering, University College London, Malet Place, London, WC1E 6BT, United Kingdom b Neonatology, Institute for Women’s Health, University College London, London, WC1E6HU, United Kingdom c Institucio Catalana de Recerca i Estudis Avançats (ICREA), Carrer de la Dispatio, Barcelona, 08015, Spain d eICFO-Institut de Ci‘encies Fot‘oniques, The Barcelona Institute of Science and Technology, Castelldefels, Barcelona, 08015, Spain *[danaibili2@gmail.com](danaibili2@gmail.com) ABSTRACT Optical biomarkers of neonatal hypoxic ischemic (HI) brain injury can offer the advantage of continuous, cot-side assessment of the degree of injury; research thus far has focused on examining different optical measured brain physiological signals and feature combinations to achieve this. To maximize the breadth of physiological characteristics being taken into consideration, a multimodal optical platform has been developed, allowing unique physiological insights into brain injury. In this paper we present an assessment of severity of injury using a state-of-the-art hybrid broadband Near Infrared Spectrometer (bNIRS) and Diffusion Correlation Spectrometer (DCS) instrument called FLORENCE with a machine learning pipeline. We demonstrate in the preclinical neonatal model (the newborn piglet) that our approach can identify different HI insult severity (controls, mild, severe) . We show that a machine learning pipeline based on k-means clustering can be used to differentiate between the controls and the HI piglets with an accuracy of 78%, the mild severity insult piglets from the severe insult piglets with an accuracy of 90% and can also differentiate the 3 piglet groups with an accuracy of 80% . So, this analytics pipeline demonstrates how optical data from multiple instruments can be processed towards markers of brain health. Keywords: HIE, NIRS, DCS, Machine Learning, Clustering, Biomarkers 1. INTRODUCTION Neonatal hypoxic ischemic encephalopathy (HIE) is caused by the lack of adequate oxygen and blood flow in the brain of the newborn infants1. HIE affects 2-3 in 1000 live births in high income countries, and manifests from the first hours of life 1,2 Pathophysiological changes following hypoxic ischaemic (HI) injury in newborn brain evolves over time, starting with a primary cellular energy failure, followed by a latent phase that lends itself as a treatment window, a third phase which is a secondary cellular energy failure and ultimately the tertiary phase lasting for months to years leading to further deranged function~~4~~. Proton

FAQs

What does the study use to classify HI brain injury severity?

It uses optical data from the FLORENCE hybrid broadband NIRS and DCS instrument, processed with a machine-learning pipeline based on k-means clustering.

How accurate is the classification in distinguishing control and HI groups?

The pipeline differentiates controls from HI piglets with 78% accuracy.

Can the method separate mild and severe HI severity levels?

Yes. It distinguishes mild severity from severe severity piglets with 90% accuracy and achieves 80% accuracy across three groups (controls, mild, severe).

Aug 5, 2026

Upload by Lucas Martin