Journal of Neuroscience Methods 190 (2010) 92–94

Contents lists available at ScienceDirect

Journal of Neuroscience Methods

journa l homepage: www.e lsev ier .com/ locate / jneumeth

Short communication

Force sensing system for automated assessment of motor performanceduring fMRI

Bill Rogers ∗, Wei Zhang, Shalini Narayana, Jack L. Lancaster, Donald A. Robin, Peter T. FoxResearch Imaging Institute, University of Texas Health Science Center, 7703 Floyd Curl Dr., San Antonio, TX 7829, United States

a r t i c l e i n f o

Article history:Received 25 February 2010Received in revised form 9 April 2010Accepted 14 April 2010

Keywords:Biomedical measurementsMagnetic resonance imaging

a b s t r a c t

Finger tapping sequences are a commonly used measure of motor learning in functional imaging studies.Subjects repeat a defined sequence of finger taps as fast as possible for a set period of time. The numberof sequences completed per unit time is the measure of performance. Assessment of speed and accuracyis generally accomplished by video recording the session then replaying in slow motion to assess rateand accuracy. This is a time consuming and error prone process. Keyboards and instrumented gloveshave also been used for task assessment though they are relatively expensive and not usually compat-ible in a magnetic resonance imaging (MRI) scanner. To address these problems, we developed a lowcost system using MRI compatible force sensitive resistors (FSR) to assess the performance during a fin-

Pressure measurement

Biomedical transducersHuman motor performance

ger sequence task. This system additionally provides information on finger coordination including timebetween sequences, intervals between taps, and tap duration. The method was validated by comparing

ith re

1

iin(2Wua

yDbWANtr

bcb

0d

the FSR system results w

. Introduction

Finger movement sequences are commonly used motor learn-ng tasks (Karni et al., 1995). These tasks are frequently used tonvestigate sensory, motor and cognitive systems in healthy andeurological subjects. The tasks range from single finger tappingGrafton et al., 1998) to sequences of finger presses (Walker et al.,003; Plewnia et al., 2006; Balas et al., 2007; Sheth et al., 2008;ilhelm et al., 2008). Finger tapping tasks are especially suitable for

se in functional magnetic resonance imaging (fMRI) applicationss they are well suited to the confined space in a scanner.

The tools for assessing finger movement sequences include anal-sis of video recordings (Balas et al., 2007; Bonzano et al., 2008;orfberger et al., 2009; Xiong et al., 2009), timed input from key-oards (Grafton et al., 1998; Walker et al., 2003; Plewnia et al., 2006;ilhelm et al., 2008), or instrumented gloves (Bove et al., 2007;

vanzino et al., 2008; Bonzano et al., 2008; Sharma et al., 2009).ot all keyboard and glove systems are suitable for MRI applica-

ions and when compatible the cost of optically based systems canun into thousands of dollars.

In the MRI environment, the high magnetic field generatedy the magnet plus the time varying field due to gradientsan adversely affect electronic instrumentation. Currents cane induced in wiring generating erroneous signals, damaging

∗ Corresponding author. Tel.: +1 210 567 8171; fax: +1 210 567 8152.E-mail address: rogers@uthscsa.edu (B. Rogers).

165-0270/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi:10.1016/j.jneumeth.2010.04.011

sults obtained by video analysis during the same session.© 2010 Elsevier B.V. All rights reserved.

voltages and potentially damaging heat. Any instrumentationcontaining ferromagnetic materials is unsuitable for use in thehigh magnetic field. In addition, the introduction of electronicinstrumentation can have an adverse effect on image acqui-sition. Any electrical conductor can produce artifacts. Radiofrequency (RF) interference from electronics can also degrade MRIimages.

The force sensing resistors (FSR) used in our system contain noferromagnetic material, operate with low voltage DC, and do nothave any coiled conductive paths. Therefore they function well inan MRI environment. FSRs have a variable resistance as a functionof pressure. As the pressure increases, the resistance decreases. Theactive area of the FSR has a substrate with a polymer thick film (PTF)layer. The substrate layer is separated from an electrode layer by athin spacer. The resistance of the PTF layer exhibits a decrease inresistance with an increase in pressure.

To assess the validity of our FSR system, finger movementdata were captured simultaneously with the FSR system and avideo camera then the results were compared. In addition, noisemeasurements were taken in the MRI environment to test the com-patibility of the system with MRI scan acquisition.

2. Materials and methods

2.1. Task for measurement of FSR system accuracy

In order to assess the validity of our FSR system for measuringfinger sequences, data were captured simultaneously with the FSR

B. Rogers et al. / Journal of Neuroscien

so

fibnfjp

i

FAw

fwsu

2

iwt1

isssWrdt

as

tM

the FSR system and the two raters was greater than 0.92. The R2

Fig. 1. Hardware connection diagram for force sensing system.

ystem and a Canon ZR500 video camera. These data were acquiredutside of the MRI environment.

The system was tested on three research subjects who per-ormed a finger tapping sequence (4-1-3-2-4) where finger 1 is thendex finger and finger 4 is the little finger. The task was performedy pressing and releasing the fingers corresponding to the sequenceumber in an exact order. FSR sensors were taped to the tips of the

our fingers and the hand rested on a hard plastic surface. The sub-ects were asked to perform the sequence repeatedly as quickly asossible for 240 s.

Informed consent was obtained in accordance with the localnstitutional review board protocols.

Each subject was measured during an initial baseline session.or the following week, the subjects practiced the sequence daily.t the follow up session after the week of practice, performanceas measured again.

Videos of six 240-s sessions of research subjects were analyzedor the number of correct sequences by two raters. The raters whoere blind to the study were asked to count the number correct

equences while reviewing video of the sessions in slow motionsing video player software on a computer.

.2. Hardware

The FSR sensors used in this system are from Interlink Electron-cs (Standard 402 FSR). They have a 12.7 mm diameter active surface

ith a 25 mm tail for the electrical contact connection. The sensi-ivity range is 0.1–10 kg/cm2 and the no load resistance is aroundM�.

The system uses four FSR sensors, one for each finger tip, notncluding the thumb. (Fig. 1) The FSR sensors are connected to aimple voltage divider circuit with a 10 k� resistor and a 5 V sourceupplied by a USB data acquisition module. When there is no pres-ure on the FSR, the voltage read by the DAQ module is about 0.1 V.

hen pressure is applied, the measured voltage increases as theesistance decreases. A voltage threshold, typically 0.5 V is used toetermine a finger press. The resistance of the FSR at this voltagehreshold is about 90 k�.

The FSR recording system is built around a low cost USB datacquisition module. The Measurement Computing 1208FS willample up to 8 single ended input at a sample rate of up to 50 kHz.

Twisted pair wiring leads from the sensors to a RF filter (Spec-rum Control RF Filter-DB15 56-715-003) to the signal panel in the

RI room then to the DAQ module in the control room.The total cost of all the components in the system is under $300.

ce Methods 190 (2010) 92–94 93

2.3. Software

The software has separate data acquisition and analysis mod-ules.

The data acquisition software was written in C++ and has asimple graphical user interface (GUI) for system setup and dataacquisition. Voltage data were acquired for the four sensors at100 Hz for the required session time. During acquisition the GUIprovides visual feedback for each channel in order to monitor sub-ject performance. The 100 Hz sensor voltage data were saved to afile for each session.

The analysis program was developed in Matlab. In additionto counting the number of correct sequences, intra-digit andintra-sequence timing information is calculated. A graphical userinterface provides user friendly operation. The analysis for a sessionis broken into four steps.

1. Load sensor voltage data for the session and a sequence file spec-ifying the finger tap sequence used for the session.

2. Threshold voltage data: The thresholding process assigns fingerpress to each voltage pulse above a set threshold. In order tocorrect for the possibility of switch bounce, a sliding windowmedian filter is an option.

3. Analyze sequences: The thresholded presses are then comparedto the reference sequence to determine the number of cor-rect sequences in the session. Additional calculations includesequence time, time between sequences and intervals betweensequence steps.

4. Output data for further analysis: There are two tab separatedoutput files. One contains the summary values calculated for theentire session. The second contains detailed timing informationfor each individual sequence.

2.4. Measurement of MRI compatibility

There are a number of issues in providing instrumentation inan MRI environment. One is to make sure that the instrumentationdoes not interfere with MRI acquisition. Another issue is to insurethat the MRI environment does not interfere with the instrumen-tation.

To determine if the FSR system interfered with MRI acquisition,several scans were run with a phantom to check for radio frequency(RF) noise and scan artifacts. Scans were taken with the FSR sensorson the phantom and with the sensors out of the field of view about30 cm away from the phantom. Scans were taken using a gradientecho pulse sequence which is sensitive to RF interference.

To check to see if MRI acquisition interfered with FSR data cap-ture, the sensors were placed in the scanner with a phantom anda typical fMRI acquisition run to acquire a background signal with-out finger tapping. The voltage data captured by the FSR system wascompared to a baseline data captured outside the MRI environment.

3. Results

Number of correct sequences detected by the FSR system wasvery similar to video analysis. (Table 1) As expected, the number ofcorrect sequences performed increased from session 1 to session2. The R2 (the square of the Pearson correlation coefficient) valuesbetween the FSR number of taps and each of the raters was greaterthan 0.99. The R2 values for number of correct sequences between

values for the percentages of correct taps were 0.39 for Rater 1 and0.20 for Rater 2.

The one session where the FSR system results varied noticeablyfrom the raters (Subject 2 Session 1) had an issue with the quality of

94 B. Rogers et al. / Journal of Neuroscience Methods 190 (2010) 92–94

Table 1Comparison of FSR system to video raters.

Number of taps Number of correct sequences Percent of correct taps

FSR Rater 1 Rater 2 FSR Rater 1 Rater 2 FSR Rater 1 Rater 2

Subj. 1 Session 1 533 544 547 94 99 99 88.2% 91.0% 90.5%Subj. 1 Session 2 861 846 851 153 151 153 88.9% 89.2% 89.9%

tfitpw

tlti

otaotw

twu

4

bm

TFFse

apWs

dMio

ptt

Subj. 2 Session 1 1272 1267 1297 207Subj. 2 Session 2 1306 1295 1304 252Subj. 3 Session 1 952 955 956 181Subj. 3 Session 2 1157 1139 1136 223

he video recording. The camera angle made it difficult to score asngers often occluded each other. Removing this session improvedhe R2 values comparing the FSR system to the two raters for theercentages of correct taps to greater than 0.85. All other R2 valuesere greater than 0.99.

There was no measurable RF noise during MRI acquisition due tohe FSR system. That is to be expected as the FSR system uses onlyow voltage DC. The RF filter removes any high frequency signalshat might be introduced via the FSR system wiring that wouldnterfere with MRI acquisition.

When the FSR sensors were placed on the phantom in the fieldf view, image artifacts were observed. If the sensors were out ofhe field of view, no artifacts were observed. During normal fMRIcquisition using the finger tapping protocol, the sensors would beut of the field of view. There was no measurable change in signalo noise ratio when the sensors were out of the field of view noras there any measurable signal drop.

There was a small but measureable amount of noise generated inhe FSR system by the fMRI scan. The amplitude of the noise signalas in the range of 0.01 V which is well below the 0.5 V generallysed as a threshold for a finger press.

. Discussion

The number of taps and number of correct sequences detectedy the FSR system showed very good correlation with the ratersaking this system a good replacement for video analysis.There have been FSR sensor failures but they are not common.

he failures appear to come from repeated bending. Initially all fourSRs were permanently connected to a wiring harness so if oneSR failed, the harness would have to be taken out of service forensor replacement. Now the FSRs have individual connectors forasy replacement.

In this system, the FSR sensors are used only as switches withny voltage above a threshold being deemed a finger press. Withroper calibration voltages could be converted to pressure values.hile not as accurate as a load cell or strain gauge, FSR sensors have

imilar properties.This system would be considered “MR Conditional” by the terms

efined by the ASTM (ASTM-International, 2008). It has no knownRI hazards within specified conditions of use. If the wiring was

mproperly looped in the scanner, there is the theoretical possibility

f heating due to induced currents.

The FSR force sensing system is well suited for the finger tap-ing motor learning task. It eliminates a great deal of tedious laborhat is involved with video analysis. The ability to use the sys-em in fMRI applications is also a benefit. This system could easily

236 249 81.4% 93.1% 96.0%249 250 96.5% 96.1% 95.9%181 189 95.1% 94.8% 98.9%219 223 96.4% 96.1% 98.2%

be adapted to any other paradigm where sequencing and timingwas important. With additional effort and calibration, the FSR sys-tem could be extended for use as a true pressure measurementsystem.

All of the major hardware components of the FSR system arelisted in this manuscript. In order to facilitate implementation ofthis system the authors are also willing to share the data acquisitionsource code.

Acknowledgements

This research was supported by Veteran’s Administration meritgrant awarded to Dr. Peter Fox. Thanks to Angelique Blackburn,Nicole Franco and Felipe Salinas for video analysis.

References

ASTM-International. ASTM F2503-08—standard practice for marking medicaldevices and other items for safety in the magnetic resonance environment. WestConshohocken, PA: ASTM-International; 2008.

Avanzino L, Bove M, Trompetto C, Tacchino A, Ogliastro C, Abruzzese G. 1-Hz repeti-tive TMS over ipsilateral motor cortex influences the performance of sequentialfinger movements of different complexity. Eur J Neurosci 2008;27(5):1285–91.

Balas M, Netser S, Giladi N, Karni A. Interference to consolidation phase gains inlearning a novel movement sequence by handwriting: dependence on later-ality and the level of experience with the written sequence. Exp Brain Res2007;180(2):237– 46.

Bonzano L, Tacchino A, Roccatagliata L, Abbruzzese G, Mancardi G, Bove M. Cal-losal contributions to simultaneous bimanual finger movements. J Neurosci2008;28(12):3227–33.

Bove M, Tacchino A, Novellino A, Trompetto C, Abbruzzese G, Ghilardi M. The effectsof rate and sequence complexity on repetitive finger movements. Brain Res2007;1153:84–91.

Dorfberger S, Adi-Japha E, Karni A. Sex differences in motor performance and motorlearning in children and adolescents: an increasing male advantage in motorlearning and consolidation phase gains. Behav Brain Res 2009;198(1):165– 71.

Grafton ST, Hazeltine E, Ivry RB. Abstract and effector-specific representations ofmotor sequences identified with PET. J Neurosci 1998;18(22):9420–8.

Karni A, Meyer G, Jezzard P, Adams MM, Turner R, Ungerleider LG. Functional MRIevidence for adult motor cortex plasticity during motor skill learning. Nature1995;377(6545):155–8.

Plewnia C, Hoppe J, Gerloff C. No effects of enhanced central norepinephrineon finger-sequence learning and attention. Psychopharmacology (Berl)2006;187(2):260–5.

Sharma N, Simmons LH, Jones PS, Day DJ, Carpenter TA, Pomeroy VM. Motor imageryafter subcortical stroke. A functional magnetic resonance imaging study. Stroke2009;40(4):1315–24.

Sheth BR, Janvelyan D, Khan M. Practice makes imperfect: restorative effects of sleepon motor learning. PLoS One 2008;3(9):e3190.

Walker MP, Brakefield T, Seidman J, Morgan A, Hobson JA, Stickgold R. Sleep and the

time course of motor skill learning. Learn Mem 2003;10(4):275–84.

Wilhelm I, Diekelmann S, Born J. Sleep in children improves memory performanceon declarative but not procedural tasks. Learn Mem 2008;15(5):373–7.

Xiong J, Ma L, Wang B, Narayana S, Duff EP, Egan GF, et al. Long-term motor traininginduced changes in regional cerebral blood flow in both task and resting states.Neuroimage 2009;45(1):75–82.