An Ultralow-Power Real-Time Machine Learning Based fNIRS Motion Artifacts Detection - IEEE Transactions on Very Large Scale Integration (VLSI) Systems

Aug 3, 2026

Upload by Noah

Machine learning modules often demand substantial processing power and memory due to iterative matrix multiplications or gradient computations, limiting their use in wearable functional near-infrared spectroscopy (fNIRS) systems. This study presents an ultralow-power, real-time motion artifact detection module for fNIRS, implemented on an FPGA. The approach achieves 97.42% classification accuracy with low FPGA resource utilization (38,354 LUTs and 6,024 flip-flops) and 0.021 W dynamic power, outperforming CPU-based SVM baselines and related SVM variants for embedded constraints.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Functional near-infrared spectroscopy and motion artifacts

Limitations of conventional offline methods

Prior machine learning approaches and hardware constraints

Proposed ultralow-power real-time detection approach

FPGA-based SVM for fNIRS

Accuracy, resource utilization, and power results

Index terms

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS 1 An Ultralow-Power Real-Time Machine Learning Based fNIRS Motion Artifacts Detection Renas Ercan, Yunjia Xia, Graduate Student Member, IEEE, Yunyi Zhao, Rui Loureiro, Member, IEEE, Shufan Yang, Senior Member, IEEE, and Hubin Zhao, Member, IEEE Abstract—Due to iterative matrix multiplications or gradient computations, machine learning modules often require a large amount of processing power and memory. As a result, they are often not feasible for use in wearable devices, which have limited processing power and memory. In this study, we propose an ultralow-power and real-time machine learning-based motion artifact detection module for functional near-infrared spectroscopy (fNIRS) systems. We achieved a high classification accuracy of 97.42%, low field-programmable gate array (FPGA) resource utilization of 38 354 lookup tables and 6024 flip-flops, as well as low power consumption of 0.021 W in dynamic power. These results outperform conventional CPU support vector machine (SVM) methods and other state-of-the-art SVM implementations. This study has demonstrated that an FPGA-based fNIRS motion artifact classifier can be exploited while meeting low power and resource constraints, which are crucial in embedded hardware systems while keeping high classification accuracy. Index Terms—Field-programmable gate array (FPGA), functional near-infrared spectroscopy (fNIRS), low power, machine learning, motion artifact detection, real time, support vector machines (SVMs). I. INTRODUCTION FUNCTIONAL near-infrared spectroscopy (fNIRS) is an emerging modality that aims to characterize cortical hemoglobin fluctuations through intensity measurements of diffusely scattered near-infrared light [1], [2] . It can help Manuscript received 25 July 2023; revised 2 November 2023 and 26 December 2023; accepted 13 January 2024 . This work was supported in part by the Department of Orthopaedics and Musculoskeletal Science; and in part by the Wellcome Trust and Engineering and Physical Sciences Research Council (EPSRC) through the WEISS Center, UCL under Grant 203145Z/16/Z. The work of Shufan Yang was supported in part by the SHED Project Royal Academy of Engineering under Grant IF2223-172 and in part by the Innovate U.K. KTP under Grant 013191 . The work of Hubin Zhao was supported in part by the Royal Society Research under Grant RGS\R2\222333 and in part by the Engineering and Physical Sciences Research Council under Grant 13171178 R00287 and Grant EP/W000679/1 .(Renas Ercan and Yunjia Xia are co-first authors.) (Corresponding author: Hubin Zhao.) Renas Ercan was with UCL, WC1E 6BT London, U.K. He is now with the Department of Physics, University of Cambridge, CB2 1TN Cambridge, U.K.(e-mail: [re378@cam.ac.uk](re378@cam.ac.uk)) . Yunjia Xia, Yunyi Zhao, and Hubin Zhao are with HUB of Intelligent Neuro-Engineering (HUBIN), Division of Surgery and Interventional Science, UCL, WC1E 6BT London, U.K. (e-mail: [yunjia.xia.18@ucl.ac.uk](yunjia.xia.18@ucl.ac.uk); [yunyi.zhao.21@ucl.ac.uk](yunyi.zhao.21@ucl.ac.uk); [hubin.zhao@ucl.ac.uk](hubin.zhao@ucl.ac.uk)). Rui Loureiro is with IOMS, Division of Surgery and Interventional Science, UCL, WC1E 6BT London, U.K. (e-mail: [r.loureiro@ucl.ac.uk](r.loureiro@ucl.ac.uk)) . Shufan Yang is with the Institute of Medical and Biological Engineering, School of Mechanical Engineering, University of Leeds, LS2 9JT Leeds, U.K., and also with UCL, WC1E 6BT London, U.K. (e-mail: [s.f.yang@leeds.ac.uk](s.f.yang@leeds.ac.uk)). Color versions of one or more figures in this article are available at [https://doi.org/10.1109/TVLSI.2024.3356161](https://doi.org/10.1109/TVLSI.2024.3356161) . Digital Object Identifier 10.1109/TVLSI.2024.3356161 neuroscientists to determine which brain regions are activat

FAQs

Why is motion artifact detection crucial in fNIRS systems?

fNIRS recordings are sensitive to participant motion, which introduces motion artifacts that can cause erroneous detection of functional cortical activity.

What approach does the study propose for artifact detection?

The study deploys support vector machine (SVM) as the machine learning method and implements it on an FPGA to satisfy wearable and embedded hardware constraints.

How do the results compare to CPU and other SVM-based methods?

The FPGA-based implementation achieves 97.42% classification accuracy and reports low FPGA resource usage and 0.021 W dynamic power, outperforming CPU SVM methods and other state-of-the-art SVM implementations.

Aug 3, 2026

Upload by Noah