using data visualization and signal processing to

Pediatric Rehabilitation, 2006, 1–13, preview article

Using data visualization and signal processing to characterize thehandwriting process

SARA ROSENBLUM1, DAN CHEVION2, & PATRICE L. (TAMAR) WEISS1

1Department of Occupational Therapy, Faculty of Social Welfare & Health Studies, University of Haifa, Haifa, Israel

and 2Department of Image and Signal Processing, IBM Research Lab in Haifa, University of Haifa, Israel

(Received 27 September 2005; accepted 20 December 2005)

AbstractIntroduction: Disturbances in handwriting legibility and speed are found among elementary school-aged children. The aimof this paper is to present a set of sophisticated analytical tools suitable for visualization and evaluation of handwritingdisturbances.Methods: Handwriting samples from 30 children, 15 proficient and 15 non-proficient handwriters, aged 8–9 years werecollected with the aid of a digitizing tablet. Temporal and spatial measures of the handwriting process dynamics based onsignal processing methods were developed and visually presented.Results: Significant differences between proficient and non-proficient handwriters were found in handwriting characteristicssuch as the standard deviations of letter width (t¼ 2.96, p¼ 0.008), letter height (t¼ 3.24, p¼ 0.005) and pen elevation(t¼ 2.91, p¼ 0.008). Significant differences were also found for the number of pen lifts (t¼ 2.27, p¼ 0.03), for the value ofthe correlation coefficients between letter length and time (t¼�6.62, p¼ 0.000) and between the actual and computednumber of words (t¼ 2.79, p¼ 0.01).Conclusions: The techniques described in this paper provide objective measures for handwriting performance presented ina way designed to help clinicians and educators visualize handwriting difficulties during clinical evaluation and intervention.Data visualization and analysis appear to enhance information concerning the spatial and temporal dynamics ofhandwriting.

Keywords: Handwriting, evaluation, digitizer, visualization

Introduction

Handwriting is a complex perceptual-motor task thatinvolves a very rapid sequencing of movements thatmust be controlled and regulated with considerableprecision [1]. Handwriting is generally consideredto be an ‘over-learned’ skill, i.e. one that needs to bepracticed repeatedly in order to achieve automaticity[2]. During the first 3 years of school, both legibilityand fluency of writing improve with continualpractice until proficiency is achieved [3,4]. Ideally,the process becomes almost automated such thatthe generation of text does not interfere with thecreative thinking process [5]. Once the skill islearned, handwriting becomes rapid, accurate andmechanical, with little need for active consciouscontrol [6], serving to increase efficiency and reduceredundancy [7].

For many children, the proficiency of their hand-writing is reflected by their ability to produce legibletext with minimal effort. However, other children fail

to reach this level of skill. It has been reported thatthe act of handwriting presents difficulties for�10–34% of elementary school-aged children[8–10]. The percentage of children with handwritingdifficulties reported depends upon the extent ofteacher awareness, as well as the type of evaluationtools that are available [11,12].

Dysgraphia or poor handwriting is a commoncomplaint among children and adults with learningdisabilities, appearing with or without other aca-demic difficulties [13–15]. Although poor handwrit-ing may occur for children who have no discernibledysfunction in other areas [16,17], it is oftenassociated with neurological, behavioural or medicalconditions including Attention Deficit HyperactiveDisorders (ADHD) [18,19], cerebellar ataxia [20],epilepsy [21] and leukaemia (as a result of cranialradiation) [22].

Currently, the majority of available handwritingevaluations are limited for a variety of reasonsincluding difficulties related to the identification

Correspondence: Sara Rosenblum, Department of Occupational Therapy, University of Haifa, Mount Carmel, Haifa, 31905 Israel. Tel: þ972 4 8240474.Fax: þ972 4 8249753. E-mail: [email protected]

ISSN 1363–8491 print/ISSN 1464–5270 online/06/00000 � 2006 Taylor & FrancisDOI: 10.1080/13638490600667964

and measurement of the major criteria that accountfor poor handwriting and practicalities of adminis-tering a comprehensive yet feasible assessmentwithin a classroom setting. They rely on subjectiveevaluation and their reliability and validity are,therefore, of concern. Moreover, they still focus onthe written product rather than on the handwritingprocess and, thus add very little to one’s under-standing of the nature and causes of writing diffi-culties (see [23] for further details).

Clinicians and educators have described the writ-ing style of children who have difficulties to includecomponents such as slow writing with many breaksin flow, an exaggerated pressure on the writing tooland surface, rapid onset of muscle fatigue anddifficulty in writing legibly [24–26].

Recent developments in data collection technol-ogy now permit the examination of a much richer setof handwriting outcome measures. With the aid ofa digitizing tablet and instrumented pen, the hand-writing of an adult or child can be monitored in real-time and stored in formats amenable to sophisticatedkinematic and kinetic analyses [16,17,27–32]. Thesedevices enable the measurement of much more thanthe simpler temporal variables that were measuredpreviously via subjective methodologies such as thenumber of characters a child is able to write in 1minute or the total time taken by a child to completea given text of a pre-defined number of characters[33,34]. Specifically, they also enable the researcherto evaluate and fully analyse handwriting in bothtemporal and spatial dimensions, all with greaterprecision that has the potential to provide insightinto the handwriting process.

In recent years, more attention has been devotedto the use of these objective methods to identify thefeatures of poor handwriting in children who havea variety of perceptual-motor and learning problems[10,27–29,35,36]. Such writing is often character-ized by variability in writing time and a lack ofcontinuity and fluency [17,32], longer pause dura-tions or elongated pause intervals between strokes[35,37,38] and a lack of consistency [30]. In aprevious study of 100 Grade 3 children using acomputerized system including a digitizer, laptopand evaluation software developed by the authors(POET [28]), it was found that poor handwritersperformed significantly worse on most of the testedtemporal and spatial variables than did their peers[27–29]. One of the most important findings wasthat poor handwriters maintained the pen above thewriting surface (In-air time/length) for significantlylarger percentages of the total writing time.Moreover, they did not simply pause In-air betweenthe writing of successive segments, letters andwords. Rather, the kinematic data demonstrated

a considerable movement of the pen above thewriting surface [29].

Handwriting involves very rapid sequencing ofmovements and, as such, is of considerable theoret-ical interest for the study of movement control.Analytic techniques developed in the fields of signalprocessing and pattern recognition are natural meth-ods for achieving a deeper understanding of bothproficient and poor handwriting [39–41].Regrettably this information has not yet beenwidely disseminated to the clinical and educationalprofessionals who work directly with people whohave handwriting difficulties and remains, in largepart, within the sphere of motor control or signalprocessing laboratories. This problem is reminiscentof the status of human locomotor investigationssome 25 years ago when sophisticated motionanalysis laboratories first presented the results ofdetailed gait analysis studies [42]. Typically, clini-cians and teachers are not sufficiently adept at theanalytic skills required to understand such resultsin the format in which they are normally presented.

In a recent editorial in Developmental Medicine &

Child Neurology, O’Hare ([43], p. 651) wrote thatthere is a ‘need to know more about the longitudinalcourse, early features, and intervention in dysgra-phia’. The work presented in this paper aims to meetthis challenge by developing and applying signalprocessing and data visualization tools for theanalysis of the dynamic features relevant to theevaluation and treatment of handwriting difficulties.These tools greatly extend the power of the previ-ously developed digitizer-based program, POET[28] that provided a modular, user-friendly environ-ment for the collection and analysis of handwritingsamples. The objectives of this study were (1) todevelop tools for computer-aided evaluation anddata visualization of handwriting difficulties and (2)to demonstrate the utility of these tools on a dataset of proficient and non-proficient handwriters.

Methods

Participants

Participants were 30 third grade children, aged 8 and9 years old, including 15 proficient handwritersand 15 non-proficient handwriters. The proficientand non-proficient handwriters were recruited fromregular public schools, used the Hebrew language astheir primary means of verbal and written commu-nication and were right-hand dominant. The parentsof all the children who participated in the study gavetheir informed consent.

The 30 participants were identified as havingproficient or non-proficient handwriting with theaid of the standardized and validated Teachers’

2 S. Rosenblum et al.

Questionnaire for Handwriting Proficiency (TQHP)[44,45], completed by their classroom teachers. Thequestionnaire was constructed from criteria selectedfrom the literature and handwriting assessmentsincluding handwriting legibility, speed, fatigue andcomplaints of pain or discomfort while writing[8,46,47]. The content validity of the TQHP wasestablished via a table of specifications compiled by10 occupational therapy clinicians and researcherswho were experienced in the assessment and treat-ment of handwriting difficulties. Inter-rater reliabil-ity for the TQHP was shown to be high (r¼ 0.89,p< 0.001) [45,48]. The teachers were asked to usethe questionnaire to evaluate the handwriting of eachstudent in comparison to what they would expecthandwriting legibility and speed for children of thatage and background.

Children with documented developmental delay,neurological deficits or physical impairment (e.g.Developmental Coordination disorders, AttentionDeficit Hyperactive Disorders, Cerebral Palsy) wereexcluded from the study. After the children wereclassified into groups of proficient vs non-proficienthandwriters according to the TQHP, one of theauthors (SR) administered the Hebrew HandwritingEvaluation (HHE) [47], a standardized, reliable andvalid handwriting assessment for the Hebrew lan-guage, to all of the children. One hundred per centagreement between these two measures was found;that is the same 30 children who were categorized asnon-proficient or proficient handwriters in accor-dance with the TQHP were also categorized as pooror proficient with the HHE.

The children with proficient handwriting werematched to the participants in the non-proficienthandwriting group, on the basis of gender, age,school and grade. For each child in the non-proficient handwriting group, a matched controlparticipant was chosen from his or her classroompeers and was taught by the same classroom teacher.Thus, there were no differences between the twogroups with respect to their age (mean¼8.7 years,SD¼ 0.27 years for the children with proficienthandwriting and mean¼ 8.6 years, SD¼ 0.35 yearsfor the children with dysgraphic handwriting) andgender ratio (three girls vs 12 boys in both groups).

Instruments

The Hebrew handwriting evaluation (HHE)

[49]. The HHE was developed in order to assessthe handwriting of children suspected of havingdifficulty with writing in Hebrew. In the currentstudy, the tool was used to examine a standardparagraph, assessing legibility via both global andanalytic measures (e.g. global legibility and numberof unrecognizable letters). The inter-rater reliability

of the HHE is r¼0.75–0.79; p< 0.001. Constructvalidity has been established by demonstratingstatistically significant differences between theperformance of children with proficient and withpoor handwriting (t¼�2.34; p¼ 0.027) [50].

POET—penmanship objective evaluation tool

[28,29]. An online computerized handwriting eval-uation tool, developed by Rosenblum et al. [28](http://research.haifa.ac.il/�rosens) with Matlabsoftware toolkits (http://www.mathworks.com/pro-ducts/matlab) was used to administer the stimuliand to collect and analyse the data. The evaluationwas developed in response to the absence of aquantitative objective handwriting tool for theHebrew language, but is suitable today for use withany other language. The computerized systemenables the gathering of spatial, temporal andpressure data while the participant is writing. Thesoftware consists of two independent parts, the datacollection program and the data analysis program.The data collection program is easy to use byclinicians and educators and is currently in use indifferent clinical centres in Israel. The data analysisprogram is under development and new analysisoptions are still being added. Clinicians collect thedigitizer data, transmit it to the tele-evaluationcentre via the Internet or on a CD and receive theanalysed data back to their clinic within a 1 weekperiod.

Participants were requested to copy a 107 char-acter paragraph onto A4-size lined paper thatwas affixed to the surface of a WACOM(404� 306� 10 mm) x� y digitizing tablet, usinga wireless electronic pen with a pressure-sensitive tip(Model GP-110). Displacement, pressure and pentip angle were sampled at 100 Hz via a 650 MHzPentium III laptop computer. The task was pre-sented visually on the screen in size 20-point Hebrewfont type Gutman Yad-Brush.

The studies described in this paper focused onwriting tasks written in the Hebrew language.Hebrew differs in several key ways from Latin-based scripts, as shown in Figure 1. Hebrew writingprogresses from right to left, successive letters areusually not connected even during script (cursive)writing and five letters have a different form whenthey are written at the end of a word. As in otherlanguages, some letters in the Hebrew alphabet areconstructed from two separate, unconnected com-ponents. This is illustrated in the first line of thesquare on the left side of Figure 1, which representsa sentence containing six words, each of which iscomposed of two-to-four letters. The first letterof both the second word (letter ‘Hay’) and the sixthword (letter ‘Aleph’) contain two unconnectedcomponents.

Using data visualization and signal processing to characterize the handwriting process 3

VisuDat. This is a suite of visualization toolsdeveloped via Mathematica toolboxes (http://www.wolfram.com/products/mathematica/index.html) and based on signal analysis and patternrecognition techniques to support the online presen-tation of complex handwriting data. These toolsfacilitate the inspection of complex data by noviceusers, helping them to discern relationships betweenvariables and to explore specific phenomena indepth.

VisuDat makes it easy to inspect the data quali-tatively as well as quantitatively. Initially, the rawdata are a structureless stream of six channelsincluding measurements of one channel for temporalmeasurements (time in seconds), two spatial chan-nels (stroke width and height in mm), one channel ofpressure measurements (in non-scaled units rangingfrom 0–526) and two channels of pen orientationmeasurements (pen elevation (0–90�) and penazimuth (0–360�)). The azimuth is a horizontalmeasure of the angular distance from the direction ofwriting on the digitizer and the elevation is a verticalmeasure of the elevation from the surface of thedigitizer.

These raw data channels have been manipulatedinto structured and meaningful entities in order toenable inspection of different handwriting measuresvia various projections of the data. For instance, oneis able to view any measure, raw or calculated, eitherfrom an On-Paper viewpoint or an In-air (time whilewriting that the pen is not in contact with the writingsurface) viewpoint. To this end, the data wereseparated into their On-Paper and In-air compo-nents. Each component was then further dividedinto strokes such that each stroke or group of strokesmay be manipulated directly and individually. Astroke is, therefore, the baseline unit that can bedescribed and analysed in terms of its specialfeatures, such as its pressure, time and orientation.In order to be able to analyse these data with greatersophistication and automation, VisuDat wasdesigned to have the capability of providing a still

higher level of data structuring. Thus, VisuDat canconstruct words and sentences from the initialstructureless lists of strokes. This gives researchersand clinicians greater power and flexibility, enablingthem to delve deeper into the process of writing.

Procedure

The study was approved by the institutional reviewboard of the Israeli Ministry of Education. Thechildren’s parents signed an informed consent form.

All participants were tested individually undersimilar environmental conditions, a quiet classroomin their school during the morning hours. Theparticipant was seated on a standard school chairand in front of a school desk, appropriate to his orher height. The paragraph copying task was pre-sented visually on the screen. The paragraph waswritten by the participant on normal writing paperwith printed lineature, which was affixed to thedigitizing tablet. The same tester carried out allcomputerized data collection sessions.

Data presentation and analysis

The visualization and processing techniques areillustrated via the presentation of the results ofanalyses of handwriting texts sampled from twochildren, a poor handwriter and a proficient hand-writer. This series of paired samples illustrates howthe two or three dimensional (3-D) monochrome orcolour visualization and processing techniques mayenhance one’s conceptualization of the dynamichandwriting process in the spatial and temporaldomains, as well as highlight differences betweenpoor and proficient handwriters. The SPSS statisti-cal package (version 10) was used to compute T-teststo compare group differences across the dependentmeasures of the handwriting product and process(e.g. letter width, number of times pen lifted frompage, number of reversals in direction of writing)for the paragraph writing task.

Figure 1. Text of a 30 word, 107 character paragraph written by a proficient 8-year-old handwriter (left) and a poor handwriterof the same age (right).


Results

The visualization and processing techniques areillustrated via the presentation of samples of hand-writing texts from two children, a poor handwriterand a proficient handwriter. Figure 1 shows a 30word, 107 character paragraph copied by a proficienthandwriter (left panel) and copied by a poor hand-writer (right panel). The novel element is that eachtext is shown embedded within an x� y grid, whichhighlights the spatial organization of each writtenpassage. One can readily observe the overall order-liness of the paragraph on the left with even spacingbetween successive letters and words and the judi-cious use of available surface space. In contrast, thecharacters and words in the text on the right aremore variable, unevenly spaced and have anomaliesin letter height and width. This information is far lessapparent when only means and standard deviationsare used to summarize the various outcome mea-sures used to analyse the text.

The presentation of the text within a grid, asshown in Figure 1, enhances one’s ability to discernthe spacing between successive letters and wordsand to identify anomalies in letter height and widthand the line slope. Moreover, it enables the readyselection of segments of text that warrant furtherstudy. For example, it is easier to notice words withinsufficient spacing between successive letters suchas the last word in the third line.

It is especially important to note that the hand-writing anomalies of the poor handwriter (rightpanel) become more apparent as the child progressesin the writing of the paragraph. It is only towards theend of the first sentence that corrections appear andthe writing veers increasingly upward and downwardfrom the line as the text progresses. Increasing theresolution of the grid would further enhance thevisibility of handwriting anomalies.

A replay of the recorded traces of such texts(which cannot be shown in the hard copy medium ofthis report) was also developed and implemented viaMathematica. This animated process enhances thevisualization of disruptions in the flow of writing andprovides important information about writing incon-sistencies and dysfluencies in the non-proficienthandwriting.

Figure 2 shows the In-air trajectories (i.e. excur-sion of the pen when it was above the writingsurface) for the same text shown in Figure 1, againfor a proficient writer (left) and for a poor writer(right). Strokes that occurred when the pen waspositioned above the tablet are marked in grey orbold black strokes, where grey strokes are the‘obligatory’ lifts (i.e. those the writer creates whengoing from one word to the next or from one lineof writing to the next) and bold black strokes are the‘unnecessary’ lifts (i.e. those the writer creates whenthe pen is lifted for no purpose related to the flow ofwriting). Closer examination of these data revealedthat the number of times that the child raised his pento a height greater than 6 mm while writing is ameaningful measure as will be described below ingreater detail. For example, the proficient writerraised his pen 28 times while writing this text,whereas the poor writer raised his pen 37 times. Thisexample illustrates the unnecessary effort expendedby some children while writing.

An analysis of pen lifts showed that proficientwriters raised the pen primarily when obligatory,that is at essential places such as the end of linesand the end of words. In contrast, non-proficientwriters frequently lifted the pen at non-essentiallocations which may have been due to poorplanning and the need to correct errors. It isimportant to note that the identification of theoccurrence and type of pen lifts was entirelyautomatic.

Figure 2. The In-air trajectories during the writing of a 30 word text by a proficient 8-year-old writer (left) and a poor writerof the same age (right).


Previous studies [28,29] have shown the impor-tance of the In-air measure for differentiatingbetween poor and proficient writers; analyses of the‘temporal dynamics’ were performed with particularemphasis on when the pen moved while In-air.In Figure 3, the results of a newly developed analysistechnique shows the distribution of these In-airsegments for a proficient (black) and poor (grey)writer when they copied a 107 character paragraph.The x-axis represents the length of In-air segmentsin milliseconds and the y-axis represents relativefrequency as a function of segment length. Thedistribution for the proficient writer is skewed tothe left of the graph, with the majority of segmentlengths being less than 100 ms and extending to nomore than 250 ms. In contrast, the poor writer hasa wider distribution of segment lengths extending toalmost 600 ms.

In order to further develop ways to documentvariability in the temporal domain of non-proficienthandwriters, as well as the spatial features of the

handwriting product, an analytic procedure wasdeveloped to identify words via the automaticgrouping of successive characters into logical units.The algorithm takes into account the relative spacingbetween the start and finish of each character andeach word. Too much space between characterstrokes will cause it to remain isolated from adjacentstrokes. Too little space between character strokeswill lead to the formation of pseudo-words, usuallycreated from the conjunction of two successivewords. This procedure enables a connecting line tobe drawn from the location of the first stroke untilthe completion of the last stroke of each unit ofwriting. As shown in the left panel of Figure 4, one isable, at a glance, to perceive the division of the textinto logical units which, for a proficient writer, arewords. Indeed, for this writer, there are neitherisolated segments nor letters associated incorrectlyinto pseudo-words. In contrast, as shown in the rightpanel of Figure 4, the logical units do not alwayscorrespond to words for the non-proficient writer.

Figure 3. The distribution of In-air segments for a proficient 8-year-old writer (black) and poor (grey) writer of the same age when theycopied a 107 character paragraph.

Figure 4. An automatic computerized division of the text (as above) written by a proficient 8-year-old handwriter (left) and a non-proficient handwriter of the same age (right).


In this case, 16 letters have been misconnected intopseudo-words (usually made up of two successivewords) and there are four isolated segments.

The outcome of this procedure provides a mea-sure of the difference between the actual numberof words in the paragraph (30) and the number ofwords computed by the algorithm. For example, inFigure 4, there is no difference in the number ofactual words and the number of computed wordsfor the proficient handwriter whereas, for the non-proficient writer, only 17 out of the 30 words areidentified correctly as words.

Based on the evidence revealed in Figures 1–4regarding the inconsistency and variability of thewriting product and process, an algorithm wasdeveloped to quantify the relationship betweentemporal and spatial variables for each writtencharacter. In Figure 5, the rectangles representthe actual width and height of each stroke whilethe circles represent the time spent in drawing thestroke. The presentation of the data in this wayhighlights incongruencies between the size of therectangles (representing the character’s spatialdimensions) and the circles (representing thecharacter’s temporal history). These incongruenciesindicate that the writer is inefficient, spending toomuch time in the formation of certain characters.

For clarity’s sake, Figure 6 shows the same resultsfor a sub-set of the data presented in Figure 5.Again, the key feature of this figure is the incongru-ent relationship between character size (as depictedby the rectangles) and time spent forming thecharacter (as depicted by the circles). For example,the poor handwriter (lower panel) spends an inor-dinate amount of time when writing final letters suchas the final zadi ðSÞ (indicated in Figures 5 and 6 bythe black arrows) relative to the size of this letter.These final letters are known to be difficult

characters in Hebrew writing due to their complexshape [51,52].

These results are summarized in Figure 7, whichdepicts individual strokes as points in the space-timeco-ordinate system. Each stroke is represented byits length in millimetres and by the time needed toproduce it in milliseconds. Observing this cloud ofpoints provides a visual overview of the correlationbetween the spatial and temporal behaviour ofproficient and non-proficient writers and to whatextent the data represented by the cloud of points areclose to a straight line fit. Of course, a correlationcoefficient for the whole paragraph containing 107characters is also calculated. However, Figure 7reflects visually. It is evident from this figure that thedistribution of characters for the poor handwriter(right) is much more dispersed than that of theproficient handwriter (left), as reflected by therespective correlation coefficient values (r¼ 0.93for the proficient and r¼ 0.65 for the poor writer).This graph also highlights the temporal and spatialvalues while writing different characters. For exam-ple, most letters written by the proficient writer(left), with the exception of six characters, are nolarger than 15 mm and require 1.00 s to write. Incontrast, the non-proficient handwriter (right) has

Figure 5. The relationship between character stroke size (height and width), represented by the rectangles and time spent in drawing eachcharacter, represented by the circles.

Figure 6. The relationship between character stroke size (heightand width), represented by the rectangles and time spent indrawing each character, represented by the circles. The sampleshown here is a small part of the text displayed in Figure 5.


written more characters that are far from the ‘bestleast squares fit’ line, i.e. five characters which arearound the 10 mm but required between 3.0–10.0 sto write.

The computerized system not only recordsthe temporal and spatial measures of the handwrit-ing process, it also records changes with which theangle of the pen is held while writing. The goal isto depict the way a person holds his/her pen. The3-D plot shown in Figure 8 represents the locationof the pen tip at the centre of a cube. The lowerplane of the cube represents the writing surface.The length of the pen is represented by two lines;the smooth line indicates its position and orienta-tion at the start of the word, whereas the stripedline represents its position and orientation at theend of the word. The blue arcs located betweenthese two lines represent the angular trajectory ofthe upper end of the pen resulting from changesin its azimuth and tilt in the x, y and z axes. Theleft panel of Figure 8, representing a sample froma proficient writer, demonstrates that this writerheld the pen tilted forward towards the paper and

made only small changes in angle while writing theword. In contrast, the sample shown in the rightpanel, which was taken from a poor writer, showsan angle of 90� between the initial position of thepen and paper as well as a greater range of penangles. This graphic visualization highlights thepen’s angular trajectory by eliminating translationalmovement of the pen from character to characterwithin the word.

These dynamic visualization techniques are capa-ble of revealing additional information on handwrit-ing difficulties when they are applied to an entiretext. For example, the differences in pen angulartrajectory between the poor and proficient hand-writers are even more apparent when the graphicalanalysis described above is applied to the completetext of the 107 character paragraph used in thisstudy. Figure 9 illustrates the curves of the pen’strajectory used in writing samples from the sameproficient writer (left) and poor writer (right) whosedata were presented above. Note how much vari-ability is apparent in the poor writers’ sample incomparison to the highly consistent trajectory used

Figure 7. The correlation between time taken to write a stroke and its size for all 107 characters of a copied paragraph.

Figure 8. The elevation and azimuth of the pen when it is in contact with the writing surface, shown while a four character word was copiedby a proficient 8-year-old writer (left) and poor writer of the same age (right).


by the proficient writer when the entire text isrepresented.

The visualization approach introduced in thisstudy has the potential to lead to a deeper under-standing of the handwriting process and appears toidentify many salient features for individual hand-writers. However, although this approach is instruc-tive in and of itself, further measures need to beidentified to enable a quantitative analyses of thehandwriting product and process. To accomplishthis, the paragraph copying tasks of the 15 proficientand 15 poor handwriters in the study sample wereanalysed with the algorithms used to produceFigures 1–9. T-tests were then used to determinewhich spatial and temporal measures differentiatedbetween poor and proficient handwriters. As pre-sented in Table I, significant differences were foundbetween proficient and non-proficient handwriters

for some of the developed measures including thestandard deviation of letter width (t¼ 2.45, p¼ 0.02)and letter height (t¼ 2.25, p¼ 0.03), the standarddeviation of pen elevation (t¼2.91, p¼ 0.001),the number of pen lifts (t¼2.27, p¼ 0.03), thedifference between actual and computed number ofwords (t¼ 2.79, p¼ 0.001) and the correlationcoefficient between letters size and performancetime (t¼�5.18, p¼ 0.001).

Discussion

Some 45 years ago, Herrick [53] remarked on thebenefits of having evaluation tools sophisticatedenough to document an individual’s handwriting,in a way that would be responsive to the variabilityof different handwriting features among typical

Figure 9. Elevation and azimuth of the pen when it in contact with the writing surface, shown while a 107 character paragraph was copiedby a proficient 8-year-old writer (left) and poor writer of the same age (right).

Table I. A comparison between proficient and non-proficient handwriters based on measures obtained via the visualization and analyticprocesses presented abo.

Proficient (n¼15) Non-proficient (n¼15)

Handwriting measures M SD M SD t df* p

Letters width (mm) 1.85 0.53 2.25 1.01 1.33 21.25 0.19Standard deviation of letter width 1.31 0.29 1.81 0.58 2.96 20.52 0.008**Letter height (mm) 3.59 1.09 4.69 2.87 1.38 17.96 0.18Standard deviation of letter height 2.09 0.37 3.04 1.07 3.24 17.33 0.005*Number of pen lifts 27.93 7.88 42.66 23.81 2.27 17.03 0.03*Difference between actual and computed number of words 5.00 4.75 11.06 6.93 2.79 24.77 0.01*Correlation coefficient 0.840 0.07 0.60 0.12 �6.62 23.01 0.000**Mean elevation 64.39 6.90 61.82 13.15 �0.67 21.16 0.51STD elevation 2.80 0.80 4.11 1.54 2.91 21.10 0.008**Mean Azimuth 87.41 54.66 93.66 56.00 0.30 27.98 0.75STD Azimuth 9.42 3.92 15.49 13.59 1.66 16.31 0.11

df* for unequal variances. *p< 0.05, **p< 0.01


children and adults. Much more recently, O’Hare[43] has called for further research to be able tobetter respond to the challenge of dysgraphia eval-uation and intervention. Throughout the interveningyears there have been regular appeals to developobjective tools feasible for use by educators andclinicians [54].

The objective of this paper was to respond to theseneeds by presenting an overview of how visualizationtools based on signal processing methods may beused to support the online presentation of complexhandwriting data. These innovative techniques,providing new options for the analysis and presen-tation of the spatial and temporal dimensions ofhandwriting, greatly extend the power of the recentlydeveloped POET digitizer-based handwriting analy-sis program [28].

First, they help to achieve a bridge betweeninformation obtained via traditional evaluation andthe more recent objective, digitizer-based data.The figures in this paper illustrate some of thespatial features that have been previously consideredin handwriting product evaluation scales, most ofwhich include criteria such as letter size and slant,spacing between letters and words, the straightnessof lines, letter forms and shapes and the generalmerit of the writing [8,26,49,55–57]. Despite theyears of study, many questions remain regardingwhich of these criteria constitute the critical compo-nents of handwriting readability and how thesecriteria can be measured [8,58–62]. The contribu-tion of the current study is in its presentation ofa variety of novel techniques in which these criteriamay be visually represented and easily measured andin the demonstration of how they may be used togain a more comprehensive appreciation of thedistinctive characteristics of the poor handwritingprocess.

Consider, for example, the criteria of letter sizeand the spacing between letters and words which aretypically included in analytic handwriting scaleevaluations. Display of the written product on ascaled grid, as shown in Figure 1, enhances inspec-tion of the phenomena and facilitates objectivemeasurement of the spaces between letters andwords in millimetres (rather than by making subjec-tive estimations as usually performed (e.g. CHESevaluation, [61] or BHK [33]). Since this techniqueis computerized, it enables an automated tabulationof objective measures for letter width, height andthe distance between letters and words (cf. Table Iand Figure 4).

Secondly, the analytical and visualization tech-niques presented in this paper provide the possibilityof presenting a readily understood display of objec-tive measures of handwriting to a child with hand-writing difficulties as well as to the parents. This

visual presentation of the data may contribute to theclient’s participation in the evaluation process. Thegraphic representations provided by the VisuDatsystem illustrate phenomena that would have beenotherwise obscure to non-expert viewers. Thus, theimplications of the variability of the child’s spatialand temporal handwriting may be demonstrated viavisual images that may contribute to the evaluationand treatment of handwriting difficulties. Forexample, the intricate trajectories of the pen andmovement reversals shown in Figures 5 and 6highlight the letters most problematic for the childand for which an inordinate amount of effort isexpended. Close examination of the individualletters that entail large variability and extraneouspen movements provide clues to therapists andeducators regarding the source of the child’s hand-writing difficulties (e.g. whether letter formationhas been sufficiently internalized or whether strokeshave been correctly produced in the correctsequence).

Much literature exists to support the effectivenessof visual and/or auditory feedback for evaluation andtreatment goals. Commonly referred to as biofeed-back, a person receives visual or auditory feedbackregarding internal processes that are normallyunconscious [61]. The benefits of the visual imagesfor evaluation and treatment among children withdifferent pathologies such as ADHD [63], migraine[64] or childhood seizure disorders [65] have beenwidely discussed. Furthermore, such visualizationtools facilitate the inspection of complex data bynovice users so that they may discern relationshipsbetween variables and explore specific phenomenain depth [66].

Bruinsma and Nieuwenhuis [55] have suggestedthat self-assessment may encourage students toimprove their handwriting and to become aware ofchanges. The extent to which the provision of visualimages highlighting handwriting difficulties is able tocontribute to a child’s participation in the evaluationand intervention process should be investigated asthis is an element of handwriting assessment thathas not been sufficiently feasible to date [23].

The contribution of visualization in other disci-plines has been recognized to nurture a meaningfuldialogue between researchers and practitioners.For example, visualization techniques have led toadvances in medical informatics research as well asin medical education and have also had significantclinical implications for application in both diagnosisand intervention [66,67]. Moreover, such visualiza-tion systems likely represent the initial stagestowards the development of more reliable and validevaluation tools, a need well recognized by research-ers familiar with the currently available handwritingassessments [59].


The visualization options presented in this paperwill likely stimulate the development of new mea-sures to assist in the identification of additionalimportant features of handwriting. For example, thedata as presented in Figure 1 have already ledresearchers to attempt to find a measure that can beused to reveal the consistency in the size of lettersand in the spacing of letters and words as they areproduced within a given text. The results shown inFigure 4 underscore the difficulty that some hand-writers have in spacing successive letters and words,and the results shown in Figures 5 and 6 highlightdiscrepancies between stroke length and time towrite.

Some of the routines developed for this paperresult in the presentation of results on the handwrit-ing process that the human eye alone is incapable ofdiscerning. For example, in a previous study, In-airtime was found to be an important variable capableof significantly distinguishing between proficient andnon-proficient handwriters [29]. In the currentstudy, the In air phenomena was further investigatedthrough the use of the newly developed analytictechniques. The spatial inconsistencies and dysflu-encies that typify the On paper pen movements ofpoor handwriters are particularly apparent when theIn-air pen trajectories are illustrated as shown inFigure 2. The difficulties that plague non-proficientwriters during their ‘non-writing’ periods seem to besimilar to those that limit their performance duringtheir ‘writing’ periods with respect to writing dys-fluencies and the increased energy requirements thatcharacterize their writing. This feature of non-proficient handwriting is also emphasized by thegraph in Figure 3, which illustrates the increaseddistribution of the lengths and frequencies of In-airsegments produced by a poor writer vs the proficientwriter. Furthermore, the computerized calculationof the number of non-obligatory pen lifts thatcharacterize the writing sample produced by a non-proficient handwriter as compared to that of aproficient handwriter (cf. Figure 2) further highlightsthe atypical writing dynamics and dysfluencies thatcharacterize non-proficient handwriting production.Indeed, the number of pen lifts was found tosignificantly distinguish between poor and proficienthandwriters.

The angular trajectory of the pen is a feature of thehandwriting process that has not previously beenexamined in detail. The dynamic visualization tech-niques illustrated in Figures 8 and 9 revealed aremarkable difference in the consistency of thepen’s elevation and azimuth when manipulated bya proficient vs a non-proficient handwriter. Thepresent results demonstrated that the non-proficienthandwriters utilized a significantly more variablerange of pen elevations due to an increased number

of translational movements made by the pen as theyprogressed from character to character while writinga word. Indeed, the range of pen angular trajectoriesseemed to be the major differentiating aspect of thepoor handwriting process, as no significant differ-ences were found between the study groups for themean elevation and azimuth of the pen when incontact with the writing surface.

Conclusions

The graphical and analytical techniques illustrated inthis paper highlight the major features that charac-terize poor handwriting including a lack of move-ment precision, dysfluency and inconsistency inspatial and temporal measures. They support thefindings of previous studies [9,19,21,26,27,31,32].These automated, language-independent visualiza-tion and analysis procedures have highlighted manysalient features of the differences between poor andproficient writing which have the potential to be veryuseful for clinicians, educators, parents and childrenin the evaluation and remediation of handwritingdeficiencies. One of the main purposes of this workwas to develop a tool that would become a usefulclinical tool. Hence, the data collection part of thesoftware has been designed to be easy to use. It hasalready been implemented in several clinical andeducational centres around the world and used indifferent languages. The clinician and teacher onlyrequires a laptop computer and x� y digitizer. Theportability of these instruments means that theevaluation can take place in the child’s naturalenvironment. Data analysis is carried out centrallyfrom data transmitted by clinicians. They, in turn,receive a full report of the child’s handwritingdifficulties.

The authors are continuing to develop additionalanalytical routines that will provide greater capabil-ities for identifying handwriting difficulties. Resultsof the type presented in this paper point to thebenefit of flexible visualization and analysis tech-niques. A future study will directly examine theresponses of clinicians and educators to these resultsand modify the data presentation in accordance withthis feedback.

Acknowledgement

The authors thank the University of Haifa’sCaesarea Edmond Benjamin de RothschildFoundation Institute for InterdisciplinaryApplications of Computer Science for seed moneyto carry out the initial stages of the study. We alsothank the Israel Science Foundation for their finan-cial support.


References

1. Conteras-Vidal JL, Teulings HL, Stelmach GE. Elderlysubjects are impaired in spatial coordination in fine motorcontrol. Acta Psychologica 1998;100:25–35.

2. Van Galen GP, Morasso PG. Neuromotor control in hand-writing and drawing: Introduction and overview. ActaPsychologica 1998;100:1–7.

3. Laszlo JI, Broderick P. Drawing and handwriting difficulties:Reasons for and remediation of dysfunction. In: Wann J,Wing JAM, Sovik N, editors. Development of graphic skills.London: Academic Press; 1991. pp 59–280.

4. Maeland AF, Karlsdottir R. Development of reading, spellingand writing skills from third to sixth grade in normal anddysgraphic school children. In: Wann J, Wing JAM, Sovik N,editors. Development of graphic skills. London: AcademicPress; 1991. pp 179–189.

5. Scardamailia M, Bereiter C, Goleman H. The role ofproduction factors in writing ability. In: Nystrand M,editor. What writers know: The language process andstructure of written discourse. New York: Academic Press;1982. pp 173–210.

6. Longstaff MG, Heath RA. A nonlinear analysis of thetemporal characteristics of handwriting. Human MovementScience 1999;4:485–524.

7. Latash ML. Automation of movements: Challenges to thenotions of the orienting reaction and memory. In: LatashML, editor. Progress in motor control. Champaign, IL:Human Kinetics; 1998. pp 51–88.

8. Rubin N, Henderson SE. Two sides of the same coin:Variation in teaching methods and failure to learn to write.Special Education Forward Trends 1982;9:17–24.

9. Smits-Engelsman BCM, Van Galen GP, Michels CGJ.Teachers assessment of the motor proficiency level andmotor writing skills of elementary school pupils. In: Smits-Engelsman BCM, editor. Theory based diagnosis of fine-motor coordination development and deficiencies usinghandwriting tasks. The Netherlands: Nijmegen; 1965.pp 33–49.

10. Smits-Engelsman BCM, Niemeijer AS, Van Galen GP. Finemotor deficiencies in children diagnosed as DCD based onpoor grapho-motor ability. Human Movement Science2001;20:161–182.

11. Hammerscmidt SL, Sudsawad P. Teachers’ survey onproblems with handwriting: Referral, evaluation andoutcomes. American Journal of Occupational Therapy2004;58:185–192.

12. Sudsawad P, Trombly CA, Henderson A, Tickle-Dengen L.The relationship between the evaluation tool of children’shandwriting (ETCH) and teachers’ perception of handwrit-ing legibility. American Journal of Occupational Therapy2001;55:518–523.

13. Cratty BJ. Clumsy child syndromes, descriptions, evaluationand remediation. Philadelphia, USA: Harwood academic;1994. pp 143–144.

14. Johnson DJ. An overview of learning disabilities:Psychoeducational perspectives. Journal of Child Neurology1995;10:S2–S8.

15. Waber DP, Bernstein JH. Repetitive graphomotor outputin L.D. and non-learning—disabled children. Therepeated patterns test. Developmental Neuropsychologia1994;10:51–65.

16. Sovik N, Arntzen O, Thygesen R. Relation of spelling andwriting in learning disabilities. Perceptual and Motor Skills1987;64:219–236.

17. Sovik N, Arntzen O, Thygesen R. Writing characteristicsof ‘normal’, dyslexic and dysgraphic children. HumanMovement Science 1987;31:171–187.

18. Tucha O, Laufkotter R, Mecklinger L, Klein H, Lange K.Handwriting of adult patients with attention deficit hyperac-tivity disorder. In: Meulenbroek RGJ, Steenbergen B,editors. Proceedings of the Tenth biennial conference of theInternational Graphonomics Society. The NetherlandsUniversity of Nijmegen: IGS Pub; 2001. pp 58–62.

19. Schoemaker M, Ketelaars CEJ, van Zonneveld M,Minederaa RB, Mulder T. Deficits in motor control pro-cesses involved in production of graphic movements ofchildren with attention–deficit–hyperactivity disorder.Developmental Medicine and Child Neurology2005;47:390–396.

20. Sanguineti V, Morasso P, Baratto L, Brichetto G,Mancardi G, Solaro C. Cerebllar ataxia: Quantitative assess-ment and cybernetic inerpretations. In: Meulenbroek RGJ,Steenbergen B, editors. Proceedings of the Tenth biennialconference of the International Graphonomics Society.The Netherlands, University of Nijmegen: IGS Pub; 2001.pp 12–17.

21. Dubois CM, Zesiger P, Perez ER, Ingvar MM, Deonna T.Acquired epilepticdysgraphia: A longitudinal study.Developmental Medicine and Child Neurology 2003;45:807–813.

22. Reinders-Messelink H, Schoemaker L, Goeken M,Briel Van den Kamp W. Handwriting and fine motorproblems after treatment of acute lymphoblastic leukemia.In: Simner ML, Leedham CG, Thomassen AJWM,editors. Handwriting and drawing research. Amstrdam:IOS; 1996. pp 215–226.

23. Rosenblum S, Weiss PL, Parush S. Product and processevaluation of handwriting difficulties: A review. EducationalPsychology Review 2003;15:41–81.

24. Benbow M. Principles and practices of teachinghandwriting. In: Henderson A, Pehoski C,editors. Hand function in the child. St. Louis, MI: Mosby;1995. pp 255–281.

25. Kaminsky S, Powers R. Remediation of handwritingdifficulties, a practical approach. Academic Therapy1981;17:19–25.

26. Mojet JW. Characteristics of the developing handwriting skillin elementary education. In: Wann J, Wing JAM, Sovik N,editors. Development of graphic skills. London: AcademicPress; 1991. pp 53–74.

27. Rosenblum S, Parush S, Weiss PL. Temporal measures ofpoor and proficient handwriters. In: Meulenbroek RGJ,Steenbergen B, editors. Proceedings of the Tenth biennialconference of the International Graphonomics Society.The Netherlands, University of Nijmegen: IGS Pub; 2001.pp 119–125.

28. Rosenblum S, Parush S, Weiss PL. Computerized temporalhandwriting characteristics of proficient and poor hand-writers. American Journal of Occupational Therapy2003;57:129–138.

29. Rosenblum S, Parush S, Weiss PL. The In Airphenomenon: Temporal and spatial correlates of thehandwriting process. Perceptual and Motor Skills2003;96:933–954.

30. Smits-Engelsman BCM, Van Galen GP. Dysgraphia inchildren: Lasting psychomotor deficiency or transient devel-opmental delay? Journal of Experimental Child Psychology1997;67:164–184.

31. Smits-Engelsman BCM, Van Galen GP, Schoemaker MM.Theory-based diagnosis and subclassification in the develop-mental coordination disorder. In: Rispens J, Van Yperen TA,Yule W, editors. Perspective on the classification of specificdevelopmental disorders. London: Academic Pub; 1998.pp 245–264.


32. Wann JP, Jones JG. Space-time invariance in handwriting.Human Movement Science 1986;5:275–296.

33. Hamstra-Bletz L, DeBie J, Den Brinker B. Conciseevaluation scale for children’s handwriting. Lisse: Swets &Zeitlinger; 1987.

34. Ziviani J, Watson-Weill A. Writing speed and legibility of7–14 year old school students using modern cursive script.Australian Occupational Therapy Journal 1998;45:59–64.

35. Schoemaker MM, Smits-Engelsman BCM. Dysgraphic chil-dren with and without a generalized motor problem:Evidence for subtypes?. In: Colla F, Masulli M, Morasso P,editors. IGS 1997 proceedings: Eigth Binnial Conference.Nijmegen: The International Graphonomics Society; 1997.pp 11–12.

36. Smits-Engelsman BCM, Van Galen GP, Portier SJ.Psychomotor aspects of poor handwriting in children. In:Simner ML, Hulstijn W, Girouard PL, editors.Contemporary issues in the forensic, developmental andnerological aspects of handwriting. Monograph of theassociation of forensic document examiners, 1. Toronto:Association of Forensic Document Examiners; 1994.pp 17–44.

37. Schoemaker MM, Shellekens JMH, Kalverboer AF,Kooistra L. Pattern drawing by clumsy children: A problemof movement control. In: Simner ML, Hulstijn W, GirouardPL, editors. Contemporary issues in the forensic, develop-mental and neurological aspects of handwriting. Monographof the association of forensic document examiners, 1.Toronto: Association of Forensic Document Examiners;1994. pp 45–64.

38. Wann J. Handwriting disturbances: Developmental trends.In: Whiting HTA, Wade MG, editors. Themes inmotor development. Boston: Martinus Nijhoff; 1987.pp 207–223.

39. Jain AK, Namboodiri A, Subrahmonia J. Structure in on-linedocuments. International Conference on Document Analysisand Recognition (ICDAR, Seattle, 10–13 September); 2001.

40. Krishna S, Nathan KSM, Beigi J, Subrahmonia GJ, Clary H,Maruyama H. Real-time on-line unconstrained handwritingrecognition using statistical methods. InternationalConference on Acoustics, Speech and Signal Processing(ICASSP 1995) International Conference on DocumentAnalysis and Recognition, Detroit, 8–12 May; 1995.

41. Subrahmonia J, Nathan KS, Perrone MP. Writer dependentrecognition of on-line unconstrained handwriting.International Conference on Acoustics, Speech and SignalProcessing (ICASSP) Atlanta, 7–10 May; 1996.

42. Halligan PW, Wade DT. Visual gait analysis: The develop-ment of a clinical assessment and scale. ClinicalRehabilitation 1998;12:107–119.

43. O’Hare A. Hands up for handwriting. DevelopmentalMedicine and Child Neurology 2004;46 651.

44. Rosenblum S, Jessel A, Adi-Japha E, Parush S, Weiss PL.Quantitative characteristics of normal and dysgraphicchildren in Hebrew script. Proceedings of the NinthBiennial Conference of the International GraphonomicsSociety. Proceedings of the Ninth Biennial Conference ofthe International Graphonomics Society. Italy: Genova;1997. pp 71–72.

45. Rosenblum S. The development and standardization ofteachers questionnaire for handwriting proficiency (TQHP).In progress, 2006.

46. Alston J. A legibility index: Can handwriting be measured?Educational Review 1983;35:237–242.

47. Cornhill H, Case-Smith J. Factors that relate to good andpoor handwriting. American Journal of OccupationalTherapy 1996;50:732–739.

48. Rosenblum S, Weiss PL, Parush S. Handwriting evaluationfor developmental dysgraphia: Process versus product.Reading and Writing 2004;17:433–458.

49. Erez N, Parush S. The Hebrew Handwriting Evaluation.2nd edn. Israel, Jerusalem: School of Occupational Therapy.Faculty of Medicine. Hebrew University of Jerusalem, Israel;1999.

50. Devash LE, Levi M, Traub R, Shapiro M. Reliabilityand validity of the Hebrew handwriting evaluation. Israel,Jerusalem: School of Occupational Therapy, Faculty ofMedicine, Hebrew University: Unpublished; 1995.

51. Share DL, Levin I. Learning to read and write in Hebrew. In:Harris M, Hatano G, editors. Learning to read and write: Across-linguistic perspective. New York: CambridgeUniversity press; 1999. pp 89–111.

52. Shah I, Hunter L. Visual management of large scale datamining projects. Pacific Symposium on Biocomputing2000;5:275–287.

53. Herrick VE. Handwriting and children’s writing. ElementaryEnglish 1960;37:248–258.

54. Karlsdottir R, Stefansson T. Problems in developingfunctional handwriting. Perceptual and Motor Skills2002;94:623–662.

55. Bruinsma C, Nieuwenhuis C. A new method for theevaluation of handwriting material. In: Wann J, Wing JAM,Sovik N, editors. Development of graphic skills. London:Academic Press; 1991. pp 41–51.

56. Freeman FN. A new handwriting scale. Elementary SchoolJournal 1959;59:218–221.

57. Hamstra-Bletz L, Blote A. A longitudinal study on dysgraphichandwriting in primary school. Journal of Learning Disability1993;26:689–699.

58. Bonny AM. Understanding and assessing handwritingdifficulties: Perspective from the literature. AustralianOccupational Therapy Journal 1992;39:7–15.

59. Daniel ME, Froude E. Reliability of occupational therapistand teacher evaluations of handwriting quality of grade 5 and6 primary school children. Australian Occupational TherapyJournal 1998;45:48–58.

60. Graham SA. Review of handwriting scales and factors thatcontribute to variability in handwriting scores. Journal ofSchool Psychology 1986;24:63–72.

61. Phelps J, Stempel L, Speck G. The children’s handwritingscale: A new diagnostic tool. Journal of Educational Research1985;79:46–50.

62. Reisman JE. Development and reliability of the researchversion of the Minessota Handwriting Test. Physical andOccupational Therapy in Pediatrics 1993;13:41–55.

63. Fuchs T, Birbaumer N, Lutzenberger W, Gruzelier JH,Kaiser J. Neuro-feedback treatment for Attention-DeficitHyperactivity Disorder in children: A comparison withmethylphenidate. Applied Psychophysiology andBiofeedback 2003;28:1–13.

64. Siniatchkin M, Hierundar A, Kropp P, Kuhnert R,Gerber WD, Stephani U. Self-regulation of slow corticalpotentials in children with migraine: An exploratorystudy. Applied Psycho physiological Biofeedback 2000;25:13–32.

65. Govender S. Childhood seizure disorders: Diagnosis andtreatment. In: Baguma PK, editor. Cross-cultural dialogue onpsychotherapy in Africa. Sovenga, South Africa: UNIN press;1999. pp 99–107.

66. Gorbis S, Hallgren RC. Visualization technology in medicaleducation. Journal of American Osteopath Association1999;4:211–214.

67. Koch S. Designing clinically useful systems: Examplesfrom medicine : density. Advances in Dental Research2003;17:65–68.


using data visualization and signal processing to

Documents