Data processing software for electrophysiological visual exploration

Computer Methods and Programs in Biomedicine, 28 (1989) 101-109

101

Elsevier CPB 00963

Section I. Methodology

Data processing software for electrophysiological visual exploration Y. Grail, J.F. Legargasson, F. Rigaudi6re a n d M. Pizzato Service de Biophysique et Mddecine Nuddaire. H~pital Lariboisidre, 75010-Paris, France

The software we have developed originated after careful study of a routine functional visual processing session. It has three objectives: (i) to have available a large library of data processing programs; (ii) to add to these electrophysiological results, automatically transferred into a file. a clinical record card, creating a personalized file for study of visual characteristics of each patient; (iii) to allow programs providing statistical studies, creation of groups based on the same clinical features and various comparisons between files or groups. The database now contains 5000 different files and the results of its use are very encouraging. Some results are presented, especially those of a study on multiple sclerosis. The goal of the operation is the creation, in the future, of an expert system. Visual database; Visual electrophysiology; Signal analysis; Cfinical file; Statistics

1. Introduction This work originated during careful study of a routine functional visual processing session. Whatever technique is used, the main phases are always the same; (i) arrival of the patient who gives the usual personal information, and is then seen by the doctor, who writes a report giving, in particular, the results of an ophthalmological check-up; (ii) the electrophysiological examination which uses various methods of signal analysis, generally resulting in graphs or numerical tables on the line-printer [1-4]; (iii) interpretation, and preparation of a report, one copy of which is sent to the doctor who asked for it and the other kept. A typical file, therefore, contained many nonstandard documents of various origins (print-out of figures, graphs and charts, hand-written notes,

Correspondence: Y. GraB, Service de Biophysique et M~decine Nucl~aire, Htpital Lariboisi&e, 2 rue Ambroise Par~, 75010Paris, France.

sundry photocopies, etc.). It was filed in this form and ease of access varied. Furthermore, a group of files was built up, often after a long and fastidious sorting process, resulting in a jumbled collection giving incomplete or incomprehensible data. For efficient use of the collected data, their standardisation and a simplified procedure became necessary.

2. Principles and methods The software we have set up, therefore, meets the following three objectives: (i) with regard to the visual electrophysiological examination (electrooculograms (EOG), electro-retinograms (ERG), visual evoked potentials (VEP) to have available a large library of data processing programs ensuring automatic memorisation of results in a condensed and sequenced form; (ii) to add to these results an 'identity data sheet', and a diagrammatical clinical record card, thus creating a personalized file for functional study of vision, suitable for print-out;

0169-2607/89/$03.50 © 1989 -Elsevier Science Publishers B.V. (Biomedical Dixfision)

102 (iii) to allow the insertion of programs providing for the build up, and the comparison between, groups and files meeting exact specifications, with the object of making statistical studies, ranging between the simple calculation of the average value of a parameter and the evaluation of efficiency of certain examination methods. The goal of the operation is the creation, progressively, of an 'expert system' including, as far as possible, all the knowledge gained in the laboratory. This software has been written in FORTRAN and functions on computers from the Digital Equipment PDP-11 series (RT-11 SJ monitor). There is always a possibility of adapting it to another machine, as long as certain specific requirements, mainly concerning the peripherals are fulfilled. These peripherals include an analog-digital converter, two diskette units and one hard disk unit, a line-printer and a graphic table. 3. Description of the system

3.1. Constitution of the database The software for this first part consists of a series of programs which are independent of each other, but all of which share a common database structure. These programs can be divided into two groups: 3.1.1. Group I The first group entails a reduction of the data and a sequenced set-up by a collection of programs which constitute the main original feature of this software. They refer directly to the gross results from the recording of a patient, without any manual intervention. Basically, we carry out three types of examination: EOG, ERG and VEP. Without going into details of the techniques and the indications given by these tests, remember that the EOG allows appreciation of the resting potential of the eye: the variations in this are often significant in cases of poisoning, particularly by medication [5], while the ERG consists of recording the electrical activity of the retina (except for ganglion cells) and is characteristically modified by many ophthalmic diseases, with the possibility of distinguishing impairment of the macula from peripheral problems [6,7].

These first two techniques produce signals which are generally faint, in which fairly simple algorithms (detection of the maxima and minima of curves, digital filtering, calculation of derivatives and detection of 'strings', etc. [8,9]) give characteristic digital data. Therefore, an EOG for each eye will give potential rest values with average lighting, after dazzle and at the end of darkness adaptation, and an ERG will give, for each stimulus and for each eye, latency and amplitude measurements of a, b and b2 waves. As for the VEPs, they reflect the activity of the brain set off by visual perception [10] and cannot be measured without some calculations, as they are normally "drowned' by background noise of the electro-encephalogram (EEG). In order to obtain the data, the EEG picked up by one or several electrodes on the scalp is amplified and sent to a computer assembly which performs the following operations. - A 'pre-process' in real time which improves the 'signal/noise ratio' by the conventional technique of signal averaging [11], completed by automatic determination of the statistical validity of the detected peaks. This amounts to calculating the confusion probability ' p ' between the top of the waves located within the limits of latency in which the VEPs can appear, and the average EEG recorded beyond the perturbation induced by the visual stimulus. After experiments on more than 400 cases, four validity classes have been defined, ranging from 'very high' ('p'<0.001) to 'low' ( ' p ' > 0.2). With ' p ' > 0.3, we believe there is no significant peak. This technique allows independence from the absolute signal amplitude variations by using the patient's own EEG, and also takes into account all the events occurring during the recording [12]. Finally, the results of this analysis of the VEP are printed out as soon as recording is completed, in the form of a table for direct reading (Fig. 1). - A series of complementary calculations made at a later stage. Among others, they include: (a) calculation of the Fourier transform and power spectrum of the signal in order either to make a comparison with the frequency bands picked up in the EEG, or to filter by suppressing certain interference frequencies and re-shaping the filtered sig-

..

m

....

N

..

o...m..o.~

~ W W ~

~ W

~I,

H

11

+

. *

o.

N

103

o.

Ld hl lU l,tll III IzJ ~ hJ I,U bJ -I -J -I Id I.U m,l IU

*'* ,.I

=.

.o

l~J

W ~ W

~ W ~ ~ W W ~

-*

0J W ~ W W W ~ . ~

~ O'l 011/'l LIJ W W W ,=J w +,,"(.it: W I-- I-- l.m

. . . m

u

z~

l+

*o

I

l

l

l

X

l

X

~ ~

~ 0 ~ * * . * . ,

X

--m I-

0,, 0 O ) 0,. " 0 P'+ ¢ ) f 9

X l

r,

t.

l

l

l

l

$.,

U

Z

Z

o.

t.

. o . o oo . . . .

o.

~

*o

~o .o

.o

oo

. . . . . .

.+

....

•

o~

¢0 U

U

Z ILl

E

l

,.,

k3 ~ ) f ~ 0 *.~ *..* Cq r l

¢[: .J ~.-

0

~"

0 ~,. W

E

W

W

-o

W

Iz,ltU ~

IJJ

,,,

tU

ILl

~

I-4 M

tU

2: tU

LLI

.,..I

~I tO l.g I~

bd h l,IJ

.",

~

U =

•

N II

~

W tU ee

tU re

'+'

W

I,,-~ I.~

W

W

~

lU U 2: IU

"J" ~I,,.

N"

LIJ m, W

--

o.

....

+.

++ ++

,4111 I

:1 ® L . . . . . . . . . .

@@

.

oo..-=

.

.

.

.

.

.

.

.

.

2~

.

~m

(/)

WWW

l,iJ m*l U.I IU I,I LiJ

W ~ W ~ Im

I

W W W W

IU LI.IW

d -+-'''' .J ~

--I bJ

.c_

w w w w ILl Im

~ D. ~ 2) l,IJIU '.LIW .J .J ..I --I

I-I

-I

0 ) I g l [/~ W W W W > -

.,¢3 IJ.l

mmmm:>

°~

~

~

j °

m

<

~'

W

td

"+

+ ~ ",0 0

I-

',~0 0

0

X

X I-Z

,~

r, 1..4

N

~2

I-,,I (.3 ..j,.

U

I,-I I~

. . . . . .

.J

U W

~

U W

1,0 tjl I-,II 0,.

0 id I-I-lU 1.0 ~:~

'~ *~ <[ ~'

0

~-~ ~dt)

~0

**

L.ql-~ ~ ttlU

~,~ f.~

Z ~ Uj .'~"

~

W 'sX I~1

l,IJ

I-

ffl. 10,. m " 0 0 I",,, +,,.'~ 0',. 0 ,. +,-I I'~ t q r ~

.J

. . . . . . . . . .

H

o.

0

0 l:Cl +,.,.

. . . . . . .

"~

Z

'~* td U

~" 0 f.1

m,* 0

W •~

.o r.

.J

GI IElU 0(.9

+-

E

l.&J

I

*-

¢.

~ W 2 : /

m

>

k"J ~ I

* lU *

O W O ~ O ~ O W

•o

I.Ll

UJ l.d W

l,d IJ

bJ

I-I

k,,4 I.,-I I,~ I " I

W

W

ILl IZ. hJ +,..

bJ ~J

.J 0 ~.

W

( ~ 0 U ) (D ¢) WO LU ( ~ Z O- ~

oo

W Im

Ld U

w w w w w :> :~ > :> :> I,,.4I-I l-I I-I I--I

ICI:

l--'I-- I--'I-- I"

hJ +," hl h llJ

.J 1:3 k*

~ ~ II

-~ ~,:

104

nal by inverse Fourier transform; (b) calculation of the intercorrelation function between two curves which allows study of the morphological similarities of the curves and their possible desynchronization, provided the physiological hypotheses are correct (for example, stimuli applied to the same patient and only differing by one carefully chosen parameter). These basics result in a program which automatically compares the conduction times concerning the left and right occipital lobes of the brain as a function of the stimulus used, and which then prints out an explicit chart of the anomalies detected [13]. All the values thus calculated are stored in a 'computer file', drawn up in the patient's name, on disk or diskette. This fairly large file easily allows conservation of the data extracted from all the recordings currently being made, and also a certain number of 'blank squares' can be reserved for any additional material obtained later. Finally, the patient's personal data are checked for accuracy by the computer and automatically dispatched. No keyboard operation is required. If new data appear in a recording, an immediate search of the existing list of patients is made and if nothing is found, the system creates a new file. As a result, it is not possible to add data to the wrong file. Furthermore, the first time the selected magnetic support material is used, a ' table of contents' is created which, in each case, provides a small space for storing simply the name and forename, the date of birth, the date of actual examination, the type or types of examination(s) carried out, and a succint clinical report which we will look at hereafter. The essential phases of the sorting programs will be included in this table of contents, following a conventional procedure which minimises the number of transfer operations between the disks, diskettes and central memory. Each case described in the contents table is then simply indexed to the main data unit of the corresponding detailed file. Of course, the interesting features of such a system are the following. - First of all, its reliability. The personal details are only put into the machine once, when the examinations begin. All data are then automati-

cally transferred directly into the various peripheral systems, with each group of data being assigned the specific coded indications for the corresponding patient thus avoiding any error or confusion between files. In addition, only the results of examinations actually carried out, and which appear on the magnetic support, are entered in code into the file. There can be no error either in the number or the nature of the examinations of a particular patient. - S e c o n d l y , its speed. The treatment algorithms, which are always the same (so that results can be compared together), are applied to the gross data and the results, in sequenced form, are transcribed into the patient's dossier as soon as the examinations are concluded. The computerized file is thus ready for quasi-instantaneous print-out, and can of course be re-opened at any time if further examination of the same patient is envisaged. Only the system itself can re-open a file when new results are being analyzed, thus avoiding once more any possibility of human error at this level.

3.1.2. Group H Tile second group entails the creation of the clini. cal file, for which two important points should be underscored: (i) it is at this moment only 1hat a considerable intervention by the doctor may prove necessary, and the conversational mode with the machine is predominant; (ii) of course, it is not a question of creating a detailed ophthalmic file, but of providing sufficient clinical facts to allow analysis of electrophysiological results, thereby giving a choice of items selected after lengthy reflection. As the program continues, the personal data can first be completed, if necessary, then a choice can be made between a 'normal patient' response on the one hand and, on the other, certain specific pathological states (not mutually exclusive) so as to obtain a primary evaluation of the case in point, even if it is imprecise. Then the screen shows, for each of the general groups above (limited, of course, to the pre-selected group or groups), a detailed table of options merely requiring a ' y e s / n o ' reply, or in some cases just a number (measurement of visual acuity, for exampie). Naturally, incompatible tests are avoided

105 (thus, a 'yes' reply to the question 'ocular fundus normal?' automatically negates any other consideration of this subject). Finally, a ' validation' case provides the option of returning to any point in the subgroup of tests which has just been completed. This system, which is very flexible in operation, minimises the time spent in front of the console and answers about 200 questions in just a few minutes. If these had to be asked one after the other, this procedure would have been tedious and would make overall control by a single operator impossible. The clinical data thus obtained can be completed by comments, for example including the World Health Organisation code for the case in point. We also found in practice that the need to give precise answers throughout the questionnaire led to greater precision in questioning and examining the patient. Finally, the system allows 100 dossiers to be entered on a diskette (8") or several thousands on a hard disk (depending on its capacity). 4. Using the database The programs in this field may ob'Aously be much more varied than in the above sector, and we do not claim to have foreseen them all exhaustively. Furthermore, we are still far from having completed the development of our various projects in this domain, but our work is sufficiently advanced for us to present the first products. Our programs are divided into three categories. 4.1. "Service' programs These give access to any data at any time. They provide print-outs (possibly in the form of tables see Fig. 1) and allow them to be corrected if required, either for altering an incorrect dossier or for total cancellation. Finally, each disk or diskette has a specific code, thus blocking access to clinical data, which cannot now be obtained in the absence of this code. -

-

4.2. "Direct-use' programs The first one allows rapid sorting of the data contained in the contents list, whether it is the

age, the sex, the dates of examinations made, their nature, or the clinical type of the affected part. Access to these functions can be of any sort, so that a pre-selection can be based on no matter what combination of criteria. At this point, the numbers of the selected dossiers will form a special group which will be treated by the other programs as an entity and dealt with wholesale, whatever the actual diversity of the corresponding physical implantations. The second takes the pre-selected files according to the criteria shown above and shows the clinical data tables on screen, such as they are defined above. After examination, the corresponding dossier is retained in the selection or removed. A qualified observer does the reading and selection in this procedure. For refining or checking the choices made using the first program, this procedure has proved far more rapid and precise than a new selection program, this time based on dozens of criteria (including parameters which are difficult to control, such as the name of the general practitioner, for example). This new selection simply replaces the preceding one but, of course, retains the same characteristics. Although not very complex from the point of view of data processing, these various direct-use programs are already very helpful in the operation of a functional visual ex~unination laboratory, particularly in the field of paperwork, because the prescribing doctor can quickly receive a dossier containing all the results in structured tabular form, pointing out the vital details. Finally, for interpretation, the immediate availability of all the results in the case in point and in similar cases, selected by simple operations, is also of considerable assistance. As the collection of data is progressively extended, a "library' diskette, containing nothing but copies of various tables of contents, is built up to allow any dossier to be found quickly and to present its main characteristics. One can also, for example, show the various synonyms, find the number of successive examinations undergone b y any one patient, etc. In our system, such a diskette can contain 11000 dossiers, representing a summary of several years of work in our laboratory.

106 H:STQGR~ES

DE P . ~ . V .

CALCUL$ $TATZSTZGUE3 EFF~CTUES ~ D£SOUE 2. CQEFF~C:ENT5 DE P O N D ~ R ~ O N : CI 0 C~ ~ 2 C3 " 3 C4 = 4

23-04-@8

REFERENCE HMtA -

DOSSIERS

RESUITATS CALCUL $UR

l

l

M

iI.iI.I l

I

:CL~SSE EN MSI I : I :

100 A 131A 161A 2~I 4 I 341A

130 160 2=0 340 400

N

: I : : I

l

100 131 161 :.~I 341 l

I

m

130 160 :.tO 340 400

a A ~I 4 ~ l

I

l

I

I

I

i

I

I

I

l

l

i

i

l

I I

N

I

l

l

i

l

I

I a.9 I 0.0 I 21.4 ; 23.8 I 10.3

I

: I I I I

II17. O. 187. 27g. 362.

CALCUL SUR

30.

DOSSZERS -

: I : : I

PZC3

I I

I

HOY : E ~ . T Y P I

2.7 0.0 3.6 3.0 2.0

: I I l

I

& white

I

I

I

l

l

N

I

0,

I

32. ;34. =. 14.

I I I

l

l

l

N [ O A l"II:'I

l

l

l

IN.PONDIP.MOY,I I

0,

! riO.

: 120. I IZ. I 3Z.

I I

l

l

l

I

l

~

l

l

l

PIOY : E C . T Y P I

I

0.0

1

! I ,' I

3.4 3,5 2.4 2.3

I t47.

0,

I

0,0

:

~.fl I I=.4 ' 2S.~ I 13.S

,

,

I 2Ib. I 303. : 373.

I I |

red & green I

P~CS POS~TZFS

I

P~C3 ~ E a ~ T Z F g

l l I

ICLASSE .~1t llS: I I I I |

I

I 60. I O. I 143. ! 79, : 6.

I I

I I I l l

l

IN.PONDIP.MQY.!

22. O. 40. 2~. 3.

REFERENC~ RVOA -

inI

: black

P~rC3 I:OS~T~FS

I I

38.

X

I

i

N

li

I I I 1 I i

i

IN.POI~OIP.MOY.:

4. 4. 33. 32. 7, i

i

I

i

I I I I | i

I

10, 10. 114. I7. 18, e

I l ; I ;

2.~ 2.5 3.~ 2.7 2,6 m

~

lilly

1 l : I I

121. 14~. 183. 27S. 3:1~.

e

i

:EC.TYP'

;

N

IN,PONOIP,MOY. :

I 2.8 I 11.7 | 9.3 I 22.1 ' 5,7

: I

O, 7,

I

I l

3~!. :~. I.

--

~

I l I I I x

l

,II.II.OI.INII

0. I=.

l

I 133. ' 12. I :3. m

l

0,0 2,t

I

I I I ~

i 0, I 14S. 1 225. I 30g. i 375;

3.8 2.4 2.9

I l

l

~

MOY :I~C.TYP,'

I

I I i i

O,O 5,= 9.8

,

I

: I

I I :6.3 I 19,0 I

I

I

I

l

itl

l | 1 I

Fig. 2. Results from staiistical calculations furnishing histograms of significant peaks (see text).

4. 3. Statistical study programs Clearly, the programs grouped under this heading depend on the personal use the individual may wish to make of his results, but, based on the file structure described above, we can envisage the following possibilities. 4.3.1. For a single dossier The responses to various stimuli can easily be studied and compared to detect any asymmetrical points, and to determine the thresholds at which responses disappear. This already allows a 'preliminary report' to be drawn up, with precise analysis of all anomalies which can be perceived from the digital values recorded. By weighting the various anomalies to a greater or lesser extent, and if necessary varying the levels of the thresholds of

detection proposed, the electrophysiologist can rapidly test one or other hypothesis, in relation to the physiopathological fa~.ts, within a conversational framework. On the other hand, he can observe the appearance of a phenomenon which, although of comparatively secondary importance, must not be overlooked. 4.3.2. For a group of dossiers This study correlates clinical information and digital values which al|ow rapid comprehension of the characteristics of a group (which is pre-determined by the capacities of the sorting programs described above). After this selection, groups are available, which are of independent value and which are stored in memory. With these, a certain number of statistical operations can be carried out

107

using various programs prepared for this purpose. For example, when analysing a VEP, for each reference number, the peaks recognized by the algorithms are stored in chronological order of appearance and according to positive or negative polarity. The operator can then split all or part of the duration of the VEP across a certain number of classes - - whether or not of equal latency, coupled or uncoupled - - within which the number of positive or negative peaks can be counted (Fig. 2 - - column 'N'). Furthermore, we consider it :,mportant that the same significance is not allocated to a series of "high-value' peaks which are clearly separated from the background noise of the EEG and to peaks which are less likely to belong to a called-up response. The program, therefore, allows us to apply a weighting coefficient C1, C2, C3, or C4, to each wave detected, representing 'low', ' average', ' high" or ' very high' probabilities. The operator can choose the values of these weighting factors. In view of the s p e ~ of the calculations, he will easily be able to test several different set-ups within a reasonable time. Thus, in the end, and within each class, an average weighted value is obtained (Fig. 2 - - column 'N.POND') which will show a certain probability level for each response. It is also possible, for example by cancelling all the coefficients except one, to obtain a simple distribution of the peaks corresponding to any statistically valid class within a chosen latency range. Finally, the program calculates and prints out the average weighting within the class in question ('P.MOY') and, within each class, the average ('MOY') and the standard deviation ('EC.TYP') for the latencies selected. An appended program immediately reproduces the results in the form of graphs (the principle is illustrated in Fig. 3). The remarkable strength of this type of program can clearly be demonstrated: it can simultaneously deal with 30 types of different stimuli in any number of files and it offers the possibility of repeating the classification of several thousands of digital data rapidly and as often as required. This has greatly helped us in our clinical research; even more as we have developed similar programs adapted to EOG and ERG recordings.

gAT( 23-04-88

o

°

38 ,DOSSIER5

mean ~-~ latency

number of Penks

o

I

;-ii

i

I,,0. I|

i/ ,

,

i

i

,

i i

I

I

I :,

t

I

I

I

.0

I

time

class I . . . 2 . . . 3

-F

Fig. 3. Graphic representation of the statistical calculations shown in Fig. 2. Time (from 0 to 400 ms) is on the x-axis and number of significant peaks on the y-axis (underlined parts of the table of Fig. 2 correspond to the first columns of the graph).

It is in this manner that we have recently been able to obtain, very rapidly, a fruitful, comparative study of a group of normal subjects and of more than 300 patients suffering from multiple sclerosis (the clinical interest of this study has been proved by previous publications [14]). This work has shown that, beyond the conventional notions of conductive delay, multiple sclerosis generally causes a weakening of the VEP waves which is mainly rooted in a rarefaction of the negative peaks, while the positive peaks are comparatively less affected. These characteristics are more marked during stimulation by checkerboards (pattern reversal - - see Fig. 4), and particularly with low contrast levels. In this way, we hope to reach a level of constructive criticism of the examination methods, to make them more efficient. Of course, the comparative study of a particular case and of the characteristics of a group, leading to a greater or lesser probability of the case belonging to the group, will soon be of great help in diagnosis. Finally, once the assimilation of a particular case within a specific group has been decided, the automatic integration of these results into the group statistics is planned, thus increasing the "knowledge" of the system and refining its future ,S~ssions. This is particularly interesting for

108

5. C o n c l u s i o n s

Normal subjects

I K~ I ~

iS : Possible M.S.

~...~x

....l,'i~T-

2[" 1~767°-

iS

,,3. C

C : Certain g.s.

Fig. 4. Mean number of significant peaks by VEP for normal subjects, possible or certain multiple sclerosis (M.S.). FL = flash stimulation, P.R. = pattern reversal. Note the weakening of the peaks (especially negative peaks) and the interest of pattern reversal stimulations (difference between possible and certain M.S. is dearly demonstrated).

analysing certain rare syndromes; we have been able to identify about 30 of them in our database, and we are actively following their characteristics in the electrophysiological recordings. 4.3.3. At group level Here, we reach the final stage in the use of our software. We have written a program allowing free creation of supplementary groups which is liable to offer regular improvement in the statistics for each group. It can also carry out operations to show the convergent or divergent tendencies between groups, for example by calculations of scores and above all the most decisive elements in the selected operational strategy. However, the number of parameters to be dealt with means that, up till now, we have only worked on simplified problems, in order to test the calculation principles selected. Finally, at all levels, access is planned to explanatory tables clearly showing the operation of the system as a function of the rules applied. Furthermore, most of these rules can be completed or modified without any need to re-write all or even part of a program (see, for example, the weighting situation mentioned above).

Thus, since the implementation of our database, which is becoming more and more complex, we have reached something which can be considered an 'expert system'. Our software has indeed both: - t h e capacity to integrate 'study' rules coming directly from the actions of scientists specialized in this field, to explain the results of their work, and, if necessary, to accept their modification by conversational methods, - a n d the capacity to improve 'knowledge' progressively, as and when new cases evolve, and to converse with a doctor, offering him a collection of logical solutions whicil largely depend on the acquired 'experience'. Of course, in this field, we are still far from having a system as efficient as we would like, and there is still a lot of work to be done. However, present research already allows us to say that: -we can operate with a reliable and precise database, which is rapidly obtained with no possibility of error, using entirely automated processing from the gathering of biological potentials to the printing-out of the patient's personalized dossier, thus considerably helping in the paperwork of a hospital laboratory; - w e can build up an 'open' calculation group, which, for each application, rapidly selects one or more reference human groups as a standard, with, if necessary, integration of the latest dossiers studied; the statistics are, therefore, automatically enriched as the data are stock-piled; - we can provide an original and powerful aid to diagnosis, by giving, with a known degree of probability, the closeness between a specific case and a given pathological group. Sometimes we can introduce new elements, in the form of an easily interpretable 'score', which the first approximate view of the dossier showed insufficiently or not at all; - we can allow groups to be compared, and the outstanding parameters to be shown clearly. In this way, we hope to be able to show the interesting features of particular investigations for discrimination between two hypothetical diagnoses, and thus to confirm the idea of an optimal strategy of visual functional exploration. The conception, the production and then the

109

installation of such a system, followed by trials and inevitable corrections, required considerable time. However, our software has now been functioning routinely for more than 3 years and our database now includes more than 5000 dossiers. Recently, we have also tried to utilise the system in accordance with the above principles, and the results are very encou_ra__~ng. Tb.is is why, although we are aware of the difficulties which fie ahead, and of the developments which our system can still undergo, we thought it might be interesting to put forward our reflections on the problem of using data in the field of visual electrophysiology, and on the solutions we have already brought into action.

References [1] R. Alfieri and P. Sole, Exploration ~lectrophysiologique de ia foncfion visuelle en photostimulafion monochromatique, Sei. M&l. 2 (1971) 157-166. [2] R. Alfieri and P. Sole, Exploration ~lectrophysio!:~iquc fonctionnelle, sectoriel!e et stratigraphique de la fonction visuelle, J. Ft. Biophys. Mbi. Nucl. 3 (1979) 63-75. [3] P. Dehon, Exploration 61ectrophysiologique de la r~tine, J. Fr. Ophtalmol. 1 (1978) 4-5.

[4] H. Spekreijse and P.A. Apkarian, Visual Pathways: Electrophysiology and Pathology (Junk, London, 1981). [5] Y. Grail, J. Keller and C. Menguy, Etude statistique de rtlectro-oculographie cfinique, J. Ft. Biophys. M&t. Nuel. 2 (1979) 77-86. [6] J.C. Hache and P. Franf~ois, Une tentative de classement des affections 61ectror~tinographiques, Bull. Soc. Ophtalmol. Fr. 6 (1976) 745-746. [7] D.A. Martin and J.R. Heckenlively, The normal electroretinogram, Doc. Ophtahn. Proc. Set. 31 (1982) 135-144. [8] J. Perrin, J. Brocas and J. Fontroger, Bases ,Sl~mentaires du traitement du signal ~ l'usage des biologistes et des mF.decins (Masson, Paris, 1976). [9] P. Pelletier, Techniques numtriques appfiques au calcul scientifique (Masson, Paris, 1971). [10] J. Desmedt, Visual Evoked Potentials in Man: New Developments (Clarendon Press, Oxford, 1977). [11] J. Max, M~thodes et techniques du traitement du signal et appfications aux mesures physiques (Masson, Paris, 1977). [12] Y. Grail, F. Rigaudiere, S. Delthil, J.F. ]~gargasson and J. Sourdille, Potentiels ~voqu~s et aeuit6 visuelle, Vision Res. 16 (1976) 1007-1012. [13] Y. Grail, J. Keller, Y. Boiteux, J.F. Legargasson and M. Pizzato, Correlation functions in the analysis of visual evoked potentials, Doc. Ophthalmol. 59 (1985) 149-155. [14] S. Mauguiere, H. Mitrou and E. Chalet, Int~r~t des potentiels ~voclu~s visuels dans la scl~rose multilocu!~re. Etude comparative des ~su!tats ,~btenus en stimulation par ~clairs lumineux et inversion de damiers. Rev. E.E.G. Neurophysinl. 9 (1979) 209-220.