Pilot study on the feasibility of acomputerized speech recognitioncharting systemFeldman CA, Stcvcii.s D: Pilot study on the tcasibility of a computerized speechrecognition charting system. Community Dent Oral Epidemiol 1990; 18: 213-5.

Abstract - The objective of this sttidy was to determine the feasibility of developingand using a voice recognition computerized charting system to record dental clinicalexamination data. More specifically, the study vvas designed to analyze the time anderror differential between the traditional examiner/recorder method (ASSISTANT)and computerized voice recognition method (VOICE). DMFS examinations wereperformed twice on 20 patients using the traditional ASSISTANT and the VOICEcharting system. A statistically significant difference was found when comparing themean ASSISTANT time of 2.69 min to the VOICE time of 3.72 min (F<0.001).No statistically significant difference was found when comparing the meanASSISTANT recording errors of 0.1 to VOICE recording errors of 0.6 (P = 0.059).90% of the patients indicated they felt comfortable with the dentist talking to acomputer and only 5% of the sample indicated they opposed VOICE. Results frotnthis pilot study indicate that a charting system utilizing voice tccognition technologycould be considered a viable alternative to traditional cxaminei/recorder methodsof clinical charting.

Cecile A. Feldman\ andDaniel Stevens^'University of Medicine and Dentistry of NewJersey, New Jersey Dentai Schooi, Newark, NewJersey, ^University of North Carolina, ChapeiiHiil, North Caroiina, USA

Key words: cotnputerized charting system;computer voice recognition; data transfer:epidemiology, dental: microcomputer

Dr. Cecile Feidman, New Jersey, DentaiSchool, Ciinical Affairs D745, 110 BergenStreet. Newark, NJ 07103. USA

Accepted for pubiication 19 January 1990

Reducing the time and personnel neededto perform dental clinical chartings cansignificantly reduce the cost of perform-ing clinical and epidemiologic oral healthresearch. This research assessed the feasi-bility of using voice recognition technol-ogy for recording clinical examinationdata by measuring time and error ratesin a pilot clinical trial.

Flistorically, exatnination data collec-tion systems for machine tabulation orcomputer processing have used citherpunch or mark-sense cards. HYAIT (1),KLEIN and colleagues (2, 3) and M()L-U";R & Pout..si;N (4) describe data collec-tion procedures using punch cards. PF.N-NELL & FIERMAN (5) Stiii-RE et al. (6),

MECKLENBURG (7), COHEN & CECII.. (8)

and DiEtiL (9) describe epidemiologiealdata collection systetns using mark-sensecards with scanners which can "sense" or"read" the marks placed on the cards. Arecorder i.s required and the data is notavailable for immediate use since thecards must be collected for "batch"processing at a later time.

This research was supported by NIH grant ..2-S07-RR-O5337-26 and llie Univetsily of Penn-sylvania School ol' Denial Medicine GCRCprogram DRR gtant /i5-M01-RR01 124.

The introduction of microcomputerdata entry systems permitted data entryinto a computer at the examination sta-tion or operatory. Data collected withthese systems may be immediately pro-cessed for analysis. PII-PI:R et al. (10)

MEADOW & ROSI:NTIIAL (11), HHLIUIRN

(12) and AciUADlo et al. (13) describekeyboard examination data entry sys-tems. To operate these systems, a record-er types clinical findings into the comput-er while the examiner recites significantfindings.

More recently other input devices suchas the X-Y digitizer described by FAIRPO

(14), the touch screen described by Dii-iiL(9), user definable keyboard or touch paddescribed by DIEIIL (9) and SOUTHARD &

RALLS (15). and voice recognition systemdescribed by BAUMGARII-N (16) have

been used. These systems were created toincrease the efficiency and accuracy ofdata entry.

Over the past 10 yr health science voicerecognition applications have grownrapidly. In the research laboratory voicerecognition is being tiscd to enter nucleicacid sequence and monoclonial antibodyresearch data into the computer (17, 18).Clinical applications in the testing stage

include a computerized speech recogni-tion system for anesthesia record keep-ing; radiology reporting: controlling sur-gical lights, microscope functions andother equipment in ophthalmic surgery:and entering organ procurement infor-mation (19 23). In addition, much re-search has been performed in utilizingvoice recognition technology to assistmotor and speech impaired people. Voicecommands can be used to operate wheel-chairs, .set environmental controls, con-trol robotic arms and access computers(22 26). Voice can improve communica-tions skills for deaf and speech impairedpatients by providing rehabilitation exer-cises (22 24).

A feasibility study to evaluate the ac-ceptability, reliability, and validity of avoice recognition telephone assessmentof the needs of chemotherapy outpatientsis reported by SIEGEL et al. (27). Results

showed tlic system to be reliable, valid,and cost efficient. Another feasibilitysttidy comparing keyboard, light pen.and voice recognition as methods of datainput fotiiid that interisi\'e care ntirsespreferred keyboard input (28). The re-searchers found keyboard input to be the"quickest, most accurate, easiest, and

214 FI:LDMAN AND STEVENS

most preferred method of entering datainto a computer" but that voice recogni-tion "may eventually provide a muchquicker and easier means of data entry."

In dentistry, voice recognition is theonly charting mechanism which can re-duce the number of personnel requiredat the examination station. If the timeneeded to perform the examination, theerror rate associated with collecting thedata, and the expense incurred during thevoice recognition procedure are less thanor equal to traditional charting systems,voice recognition would be a useful alter-native.

Materials and methods

During the fall of 1986 funding was re-ceived to develop a specialized dentalcharting system which would test the fea-sibility of voice recognition in a clinicalenvironment. A Decayed, Missing, andFilled Surface (DMFS) charting systemwas developed on an IBM PC equippedwith a KEY TRONICS KB 5152V voicerecognition keyboard. The KEY TRON-IC'S system was chosen because of itsmodest cost (US.'E850.00) and ease ofmoving the voice recognition hardwarefrom one machine to another. Most voicerecognition systems require the installa-tion of a board (or card) into the systemunit of the computer. The KEY TRON-ICS system only required replacing thestandard IBM-PC keyboard with theKEY TRONICS keyboard. This allowedthe researchers to use the voice recogni-tion system on more than one PC. At-tached to the keyboard is a headset, sim-ilar to a telephone operator headset, anda foot pedal which, when pressed, allowswords to be spoken into the computer.

The DMFS charting system was pro-grammed in the C language. Softwarewhich accompanied the KEY TRONICSkeyboard was used to control the voicerecognition hardware.

After completion of program develop-ment, 20 patients with at least 20 teethand three restorations were chartedtwice, once using the traditional record-cr/cxainitier (ASSISTANT) method andonce using the voice recognition(VOICE) charting system. During bothchartings the time necessary to completethe examination along with the nuinberol errors was recorded. Only those errorswhich had gone unnoticed by the exam-iner and/or primary recorder were writ-

lahle I. Time and error rates for VOICE vs. ASSISTANT leeotder DMFS eluttling

Time (min)ASSISTANT VOICE

Error rateASSISTANT VOICE

Mean.S.d.TesI StatisticP value

2.691.16

Paired /-testO.tKK)

3.721.08

0.10O.."? 1

Wilcoxon Signed0.0. 9

0.601.00

Rank Test

ten down as an error. This methodologysimulated a real clinical environmentwhere recorders and examiners converseto determine the correct charting.

During the ASSISTANT charting a se-cond recorder watched the primary re-corder to see if any errors were beingwritten down. During the VOICE chart-ing, the second recorder watched thecomputer screen as it registered what thedentist was saying. If the examiner notedany errors when using voice recognition,the examiner simply recharted the tooth.By recharting the tooth erroneous datais erased and the correct data saved.

Before each patient was dismissed thepatient was asked to fill out a question-naire designed to determine his or heracceptance of the voice recognitioncharting system. Patients were asked ifthey felt uncomfortable with the dentisttalking to a computer, if they felt uncom-fortable interrupting a dentist because hewas talking to a computer, and whetheror not they would go to a dentist whouses a voice charting system.

Results

The mean time for the ASSISTANTDMFS charting was 2.69 min comparedto 3.72 min for the VOICE DMFS chart-ing. There were O.I errors per chartingwhen the decayed, missing, and filled sur-face count was done with an assistantversus 0.6 errors per charting when theexamination was performed with thevoice recognition computer hardware.

The paired t-test revealed a statistically

significant difference between VOICEand ASSISTANT times (P<O.OOI) butthe Wilcoxon Signed Rank Test revealedno statistically significant difference be-tween error rates of the two chartingmethods (Table 1).

The questionnaire revealed that only5% of the participants felt uncomfort-able during the ASSISTANT charting be-cause the dentist was talking to the assis-tant, while 10% felt uncomfortable dur-ing the VOICE charting because thedentist was talking to the computer.Twenty-five percent of the participantsfelt unable to interrupt the ASSISTANTcharting while 50% felt unable to inter-rupt the dentist while the VOICE chart-ing was performed. Ninety-five percentof the participants indicated they wouldgo to a dentist who used a computerizedvoice recognition charting system (Table2).

Discussion

Results from this pilot clinical trial indi-cate that chairside time necessary to per-form a cotnputerized voice recognitionexamination is statistically longer thanthe traditional assistant recorder method.However, total time Ibr the VOICFmethod equals time the examiner spendswith the patient while total time for theASSISTANT method includes time anassistant spends with the patient, time anexaminer spends with the patient, andtime it takes to transfer data from thepaper chart to the computer. Once timefor a recorder and data transfer person-

Table 2. irc te.sponses

Charting inelhodASSISTANT VOICE

Did you feel uncomfortable during the charting procedute because Yesthe dentist was talking to the assistant/compulei? No

Did you feel that you were unable to talk to the dentist during Yesthe charting procedure? No

Would you go lo a dentist who uses a computerized voice charting YessyslemV No

n=\/(-= 19

n = 5n=l5

Notapplieable

//n

It

n

nn

1

- 1 8

= 10--10

= 19= 1

Computer voice recognition 215

nel are taken into account, the computer-ized voice recognition exam possesses thetime advantage.

Two major types of errors occurredwhile using the voice recognition system.First, the computer did not always recog-nize words it was trained to understand.For example, even though the computerhad been trained to recognize the word"next", meaning next tooth, the comput-er occasional did not indicate with a beepthat it understood the examiner. In theseinstances it was necessary to repeat theword or use keyboard entry. (This errorwas not recorded in the error rate indi-cated earlier because the errors were no-ticed and corrected by the operator with-out auxiliary personnel intervention.)The time needed to correct these errorspartly accounts for the time differencebetween the ASSISTANT and VOICEmethods.

The second type of error involved thecomputer recognizing the word the ex-aminer called into the microphone as adifferent word. For example, if the exam-iner said "mesial", the computer mightrecognize the word as being "distal".Since the computer recognized the incor-rect word as a legitimate word in its dic-tionary, the computer beeped the exam-iner, indicating that a tooth surface wasunderstood and recorded. If the examin-er did not look up at the computer screenthe next series of words would be re-corded as the charting for the distal ofthe tooth instead of the mesial surface.It is these errors which most concernedthe researchers. It should be noted, how-ever, that word recognition errors can beminimized by using a larger voice patternprint and selecting vocabulary wordswhich arc phonetically different.

Although the higher number of errorsfor the VOICE examination method wasnot statistically significant, the research-ers did not feel this error rate was accept-able. The small sample size probably ac-counted for the difference being non-sig-nificant. Future research must bededicated towards reducing the errorrate.

Patients generally accepted the VOICEcharting method even though more studyparticipants indicated that they could notinterrupt the examiner when the voicecharting procedure was performed. The

number of participants who felt theycould not interrupt the examiner whenperforming a VOICE charting could havebeen reduced if it was explained to theparticipants that the examiner could dis-engage the voice system by releasing thefoot pedal. Only one participant saidthey would not go to a dentist who used acomputerized voice recognition chartingsystem.

Additional research must be conduct-ed on a larger sample to assess time, ex-aminer differences, error rates, and re-training requirements, but results fromthis pilot study suggest that cotnputer-ized voice recognition clinical chartingsystems can be a viable alternative totraditional examiner/recorder methodsof charting. For researchers, the voicerecognition system allows data to be con-veniently recorded and available for im-mediate computer analysis.

References

1. HYATT TP. How dental statistics ate se-cured in Ihe Metropolitan Life InsuranceCompany. J Dent Res 1929; 9: 411-45.

2. KLEIN H . KRAMER MA. Dental examina-tion reeotd tbrm for statistical machinetabulation. ./ Dent Res 1948; 27: 17t)-4.

3. KLEIN H . PALMER CE. Studies on dentalcaries. X. A procedure Ibr the recordingand statistical processing of dental exami-nation findings. ./ Dent Res 1940; 19:243-56.

4. MOLLER IJ. PouLSEN S. A Standardizedsystem for diagnosing, reeording and ana-lyzing denial caries dala. Seand J DenlRes 1973; 81: 1-11.

5. PENNELL Ell. HERMAN SS. A mark-.sensecard for recording dental examinationfindings. ./ Dem Res 1952; 31: 113- S.

6. SHII-RE 1"'R, ALMAN JE, FOGELS HR.Method of recording clinical caries exatni-nations using the mark-sense system. ./Dent Res 1962; 41: 815-20.

7. MECKLENBURG RE. A dala system fordental program administration. ./ AmPent A.s.soe 1970; 80: 601-9.

8. COHEN ME, CEC tL JC. Method Ibr auto-mated entry of a dental chart schematicusing a computer card reader. CommutiityDent Oral Epidemiol 1982; 10: 264-7.

9. DtEHL MC. Automated dental epidemiol-ogy systems. III. Data system design. Na-val Dental Research Institute progress re-port 1984; NDRI-PR 84-07. 4.

10. Pit;PER K. KLAR R . KISSLER P. Use of amicrocompuler Ibr tecording dental epi-demiologic data. Communilv Dent OralEpidemioi 1981; 9: 178-81.

11. MEADOW D. ROSE.NTHAL M . A corpora-tion-based computerized pteventive den-

tistry program. ./ Am Dent Assoe 1983;106: 467 70.

12. HEI.BURN RL. Tlie computer-assisted peri-odontal chart; A tool lo aid in the conttolof periodontal disease. Compend ConlliiEdue Dent 1984; .\- 80-5.

13. AGUADtO G. PtlAlO GP. BAKlOLliCCI EG.Computerized ebarting of probing depths../ Periodontoi 1985; . 6.- 776-9.

14. FAtRPO CG. A new method ofdireet com-puter entry Ibr denial epidetniologicaldala using an X-Y' digiliset. CommunityDent OtaiEpidemiol 1979; 7. 161 4.

15. SoLiTHARD TE, RAl.t.s SA. A comptiler-ized dental examitiation record systetn.Compul Biol Med 1986; 16: 59 67.

16. BAUMC;ARTEN HS. A voice-input cotnpu-terized dental examination syslem usinghigh tesolulion graphics. Compend ContinEdue Dent 1988^ 9: 446-54.

17. ARNOLD J. EcKi:NR(it>i; VK. Lt:MKt K.PHILLIPS GJ. ScHAr.Fit;R SW. .\ compte-hensive package tbr DNA sequence anal-ysis in F O R T ' R A N IV tbr the PDP-11.Nueleie Aeids Res 1986; 14: 239-54.

18. LoNCi WJ. SHMtHLUK KE. GRUVER PE.McGutRE WR, PAt,()MHO A. EMtNi EA. Avoice-entry dala tiiangeiiient syslem: ap-plication in tnonoclonal antibody re-search. Comput Methods Proi^ramms Bio-med 1986; 23: 211 6.

19. SARNAT AJ. Compulerized speech te-eognition lor anesthesia recordkeeping.Mi'd lustrum 198.i; 17: 25-7.

20. MATtiMOTO T. IlNt!MA T A . T A t t;NO Y.IKE.HIRA 11. YAMASAKt T. FllRl'IItSA K .TsUNt^MOTO 11. StttSIIttX) I . K t l i O ^'. 1 N -

AMURA K. Automatic tadiologic reportingsystem usina speech teeogtiition. :\tedProgr Teehnol 19S7; /:.' 24.V57.

21. LE,I:MING BW. PROTER D. .IACKSON JD.Bi.t;iCH HL. SIMON M . Computerized la-diolocic reporting wilh voice datii-enlry.Radiology 1981; 138: 585 8.

22. MAN(.iOLD H. Prineiples of automatic pro-cessing of speeeli signals and their applica-tion in medical leehnology and for aidsfor handicapped. .Med Progr Techuol1988; 14: 39-56.

23. AKI:KS GA. Using your voiee: speeeh re-cognilion technology in medicine and sur-gei y. Clin Plast Surg 1986; 13: 509-1 1.

24. ERii;t5-OKi:N M. Voice recognilion devieeas a eompuler interface for motor andspeech itnpaited people. .In-h Phys MeilRehabii 1985; 65: 678-81.

25. RASTGAR MS, DEVANEV M.i. Speech le-eognition tor wheelchair eonttol. BlomcdSet Instrum I9S3; 19: 109 10.

26. SrilAi.K TB. A voice-aclivaled wheelchair.Speech Teehnol 1987; Oct/Nov; 44-8.

27. SiiXiiT K. MESAGNO FP. CHEN JY. KIT.INL. BowLt;s ME. Me KENNA M . PHVSI:R S.CriRtST G. Computerized telephone as-.sessment of the "eonetete" needs ot che-motherapy oulpalienls: a feasibility study../ Clin Oneol 19S8; 6: 1760 7.

28. KENNY GNC. Ml RCHIE CJ. Compatisonof keyboard, lighl pen and voice tecogni-tion as methods of data input. Im J ClinMonlt Comput 1988; .V 243-6.