+ +

SCORE Analyzer

Detection of subclinical Keratoconus and ectasia with software preventing SCORE Analyzer.

 

The Score software is a computer program in artificial intelligence intended to provide assistance to the clinical decision in refractive surgery for early detection of new forms of Keratoconus infra-clinic and prevention of corneal ectasia. Designed by Dr Damien Gatinel and Alain Saad between 2009 and 2011, it has obtained the CE in December 2013. It follows from the findings of a clinical research work started in 2002, and uses the data provided by the topographer Orsbcan IIz (Bausch & Lomb, Technolas PV, Germany).

SCORE Analyzer software

Delivery of the first copy of the software SCORE to its designers, Dr. Alain Saad and Damien Gatinel, January 31, 2013, the Rothschild Foundation

 

Presentation of the SCORE Analyzer software:

SCORE Analyzer software is a program of clinical decision support. A website devoted to it (in English): www.score-Analyzer.com

The software uses the data acquired by the specular topographer and elevation Orbscan IIz. SCORE software takes the classic view 'Quadmap' (4 maps) and complete it by new quantitative evidence gathered in a maps designed for a quick and easy interpretation.

Score Analyzer

Software SCORE display: the map takes the classic layout "Quaf Maps" of the Orbscan. The value of the SCORE corresponds to the calculation of a discriminant function based on 1 indices. The "zero" value is the optimal threshold. In this example, the calculated value (Score = 2.1) towards the presence of subclinical Keratoconus. This map (left eye) is that of a patient with advanced eye Keratoconus right. Despite a negative detection based on single specular topography (OPD Scan III, indices Klyce Maeda, software Corneal Navigator), the value of the Score is largely positive. The calculation of the Score is based on the use of 12 specular topographic indices, of elevation and pachymetriques (thickness).

SCORE Analyzer must be used for early detection of forms beginner subclinical of Keratoconus in myopic subjects, never operated in refractive surgery, less than 50 years of age: these characteristics derive from those eyes included in the control group used for the development of the software.

The value of the Score

The value of the score is the result of a discriminant function with 12 variables whose optimal threshold for sensitive and specific detection has adjusted to match the value 'zero '.

Score analyzer

The Score is a number whose value reflects the degree of resemblance to form fruste of Keratoconus. It comes from the collection of 12 combined linearly scales and a constant for the most discriminating threshold value to be set to "zero."

-a negative value (score < 0) corresponds to a cornea judged free form fruste of Keratoconus.

-a positive value (score > 0) corresponds to a cornea that is considered as infringement of a form fruste of Keratoconus.

The absolute value of this value (its "distance" with the threshold) is indicative of the character "normal" or "abnormal" analyzed cornea. The realization of corneal refractive surgery type LASIK is deprecated in case of positive value.

 

map "Radar."

The radar is a map with the 6 topographic indices among the most discriminating used for the calculation of the score. The values of each index are brought on a demi-axe, and the 6 semi-axles linked together so that the coordinate points connected to each of the indices form a polygon. A colour scale ranging from green to red allows a quick visual analysis of the results. The yellow color is the limit value (average more or less two standard from a normal population deviations). Orange and red colors correspond to abnormal values.

map radar

Radar map: it allows to visualize the position of certain indices used in the score function: the presence of an abnormally high value is indicated by the use of a warm color (yellow, orange, red): in this example, the vertical shift of the finest point and the value of the index I-S have values located approximately 2 gap types beyond the average.

 

The variables used in the construction of the radar are:

–          L’thickness of the thinnest point (microns) that corresponds to the minimum thickness of the corneal wall (using an acoustic factor of 0.92 for the Orbscan)

–          L’after maximum elevation (microns) which corresponds to the maximum distance in the Central 3mm between the corneal posterior data and calculated later reference sphere.

–          L’irregularity in the Central 3mm of the anterior (dioptres) which is proportional to the standard deviation of the values taken by the local curvature of the anterior surface.

–          Le vertical shift of the point the thinnest point (mm) and which is moving from the finest to the geometric center of the cornea.

–          La average difference between the Central thickness (area Central diameter 2 mm) and the thinnest thickness.

-The value of the Unlike I-S (Dioptres), which corresponds to the difference between the values average IOL 5 equidistant points located 1.5 mm of the vertex at the level of the anterior corneal surface top (S) and lower (I). It reflects the degree of vertical asymmetry of the corneal surface earlier.

 

Thick curves (product): average thickness and thinning to the thinnest point rate (speed).

The curve of medium thickness ("averaged pachymetry") consists of the average values of the thickness of the cornea from the periphery to the Center (point the thinner), according to 9 concentric rings of width 0.5 mm, contiguous to the periphery (diameter 9 mm) to the center of the cornea (where the reporting average value is that of a central lozenge of 1 mm wide centred on the thinnest point).

The thinning rate curve corresponds to the speed of the corneal wall thinning towards the end: each value is expressed as a percentage, which reflects the reduction in thickness between two adjacent rings.

thickness profile curves

The average product map (Averaged Nathan.) corresponds to the average value of the corneal thickness within 9 (width 0. 5 mm) concentric rings distributed the finer point until the periphery of the cornea. The rate of thinning map (Nathan. Thinning rate), expressed as a percentage, represents the rate of thinning from the periphery to the center of the cornea. This faster (low curve on the graph), the presence of a Keratoconus subclinical form should be suspected. In this example, it is remarkable that despite a thickness substantially higher than the average on each concentric ring, the thinning rate is substantially higher and is located on the edge of the strip of normality.

 

Origin of the SCORE Analyzer

Choosing a relevant clinical model

Keratoconus frustes forms are defined by some authors as of the very beginner infra-clinical forms, where the achievement of the anterior face of the cornea is not sufficiently pronounced it rising suspicion; the values taken by the indices used for testing below the detection thresholds for these.

How to study the topographical characteristics of the eyes with these forms frustes of Keratoconus? In theory, we could analyze retrospectively the ancient topographical maps, performed at the early stage of affection for eyes today suffering from Keratoconus turned out, in the hope to find the beginnings of this affection. Unfortunately, it is rare that patients have received a corneal topography, then they were visually asymptomatic, and showed no sign of appeal to review biomicroscopique (slit lamp).

Another clinical model can be used to track down the most beginner forms of Keratoconus: eyes are less affected patients achieved "very asymmetrical Keratoconus between the eyes., IE advanced (clinical and/or topographic form proved) over one eye, and very beginner on the other (i.e. objectively subclinical, with) negativity of detection based on topography specular Placido mode's clues). The least reached eye was included within the Group of interest of our clinical model, if from testing of the specular topography (ex: Klyce - Maeda criteria) were negative, meaning that the automatic classification by neural network (Corneal Navigator, surveyor OPDScan, Nidek, Japan) provided similar percentages strictly Dummies for the objective topographic diagnosis of Keratoconus turned out suspect Keratoconus and pellucid marginal degeneration,.. .so that the other eye was of course tested positive for the diagnosis of Keratoconus turned out.

model of the score

Patient with Keratoconus turned out of the left eye, and a cornea deemed topographically normal by the neural network, according to the calculated indices of Klyce Maeda (specular topography of Placido by OPDscan, Nidek). The right eye is asymptomatic (best corrected Visual acuity: 12/10). At the interview, there is the notion of atopic field, but the absence of eye rubbing digital repeated. The existence of an advanced Keratoconus on the left side to consider the cornea of the eye as presenting an infra-clinic form of Keratoconus. This subclinical form is necessarily a fruste form according to the definitions in force at present, as considered normal based on objective analysis in videotopographie automated Placido.

This clinical model is valuable because it allows you to establish a group of eyes which it is certain (through the achievement of the other eye) they correspond to a fledgling form of Keratoconus () subclinical1). The existence of strictly unilateral forms of Keratoconus is denied by most of the authors)2,3) It may be considered more relevant to include these diagnostic eyes as ' normal»(par erreur, ou plutôt car à un stade très débutant – inférieur àle seuil retenus pour les critères Placido), eyes which would be directly diagnosed as 'Keratoconus suspect' with Placido indices. From the moment the corneas are labeled as "suspicious", they need to reconsider the indication of corneal refractive surgery. In this context, the problem is rather to check that this suspicion is based, and not driven by excess (false positivity for indices, whose specificity is not absolute).

Eyes that have been included in our model are similar to those that pose a diagnostic problem in the check prior to refractive surgery, like the one shown here:

corneal ectasia at-risk

Patient 1 is reached a form whose expression between the right and left eyes is very asymmetric. This patient has a proven form of Keratoconus in left eye, as evidenced by the appearance of the topographic map and values calcuiees for indices Klyce and Maeda. On the right, the automated diagnosis with neural network Corneal Navigator (Nidek) from the axial specular map (OPDscan, Nidek topographer) is negative for the proven Keratoconus (KC) and suspect (KCS), the patient is asymptomatic (best corrected Visual acuity: 12/10). There is a slight asymmetry marked by an area a more arched nasal inferior, but it is sufficiently small so that all of the indices calculated are normal.
2 patient is asymptomatic, and right or left a slight increase of the lower nasal arch. Right, all indices are normal, left, only the OSI is suspect, but the neural network class both corneas of this patient as normal (NRM). The topographic map of patient 2's right eye looks a lot like the topographic map of the right eye of the patient 1. If we can issue a doubt when the possibility of a form (subclinical beginner) fruste of Keratoconus for corneas in patient 2, arguably still the presence of this condition at the level of the right cornea of the patient 1. We can define and test other indices of screening for these forms frustes by collecting data from the elevation and the total product as the Orbscan for this type of corneas.

The risk of ectasia eyes present some minor, but located anomalies within the limits of conventional topographic detection)4). The use of a clinical model based on the study of the eye the least reached in the case of a very asymmetrical breach between the eyes was used or taken over by many authors)5-8). Due to the scarcity of the apparently unilateral forms of Keratoconus)2,3), the formation of enough with eyes in some way of a form fruste of Keratoconus (negative for tests in force while their contralateral eyes are achieved) represents a certain difficulty: it took us almost 10 years to collect a large enough sample.

 

 Using a discriminant function

 

Corneas "Normal Placido" but belonging to a patient genetically Keratoconus (reached proven of the contralateral eye) have been studied in topography of elevation by scanning by light slots (Orbscan IIz, Technolas PV, Germany), and compared to a control group of healthy corneas belonging to some myopic patients operated of LASIK have reported no complications during the 5 years of postoperative follow-up.

In these eyes with Keratoconus frustes forms, many indices calculated from the elevation anterior, posterior and the product were statistically different from those measured at eyes normal)1). The irregularity of curvature of the anterior face calculated in the Central 3 mm was also statistically increased (0.98 + / 0.34 vs 1.25+/-0.38 D). However, the comparison of distributions between two samples (corneas normal vs. form fruste of Keratoconus) reveals a significant overlap in the distributions of these values.

distribution parameters of irregularity Central 3 mm

Comparison of the distributions of values measured for the irregularity of the anterior face of the cornea (area: 3 mm) between a group of affected eyes to form fruste of Keratoconus (FFKC: negative test for outcomes Klyce and Maeda, advanced eye adelphe with Keratoconus). The average value (Red Cross) is significantly more important in the FFKC group than in the control group. Respective median values (almost identical), and the superposition of the respective values of each sample strongly suggest that a test limited to the use of this index would not achieve a sufficiently discriminating test.

It is interesting to note that despite the normality of indices Placido, a statistical test that is performed between the Group of affected corneas of Keratoconus fruste and normal corneas reveals a statistically significant difference in anterior curvature. The normality of the indices Placido simply means that the values taken by the indices calculated for each of these corneas are below the thresholds initially chosen for these tests!  In absolute terms, the irregularity of the anterior face of the affected fruste Keratoconus corneas remains statistically more important than those of the normal corneas averaged: However, the superposition of the values taken by the corneas of each sample does not bode of obtaining a good sensitivity and specificity with a test based solely on that criterion, and a threshold is set lower. One could certainly identify shapes frustes of Keratoconus with this simple test and a lowered threshold, but they accuse of many normal eyes wrongly...

Other parameters were statistically significant differences between the two groups: Central average thickness was reduced (524.3 +/-35 vs 554.6 +/-36 microns), the difference between this value and the end point (point windows) was increased (12.1 +/-5.6 vs 6.8+/-3.1 microns), the distance with the geometric center of the fine was increased (0.95 + / 0.37 vs 0.64 +/-0.32) , and the elevation value later with respect to the sphere of reference calculated after centering on the thinnest point was greater next to this point (26.3 + / 11 microns vs 19.7+/-8.6 microns).

As noted previously, the averages calculated for each of these parameters, although significantly different statistically, were too close to the distribution of the measured values to enable them, considered in isolation, be really clinically significant, and can build a sufficiently sensitive and specific screening test. Yet, the gradient of thinning of the cornea from the periphery of the cornea to the Center, the index of irregularity topographic to 3 mm, the shift of the vertical of the end point, the difference between the Central Product and the finest product as well as the difference between the lower keratometry average and the higher average keratometry measured 5 mm from the vertex are among the clues showed a potential discriminatory power. However, using isolated from each index does not differentiate with sufficient sensitivity and specificity the normal corneas of the corneas of the group studied due to the overlap important values made by the sampled corneas.

On the other hand, thecombined use of these criteriabased on the calculation of a composite index with a technique ofdiscriminant analysisled to a more sensitive and specific diagnostic test. The control group used to build the test consisted of eyes of myopic patients aged under 50, made of myopic LASIK without complication (especially the ectasia), and followed for 5 years at least. Remember that the study group was composed of classified as normal eyes with a topography of Placido and neural network based on the use of indices Klyce and Maeda: these eyes are however likely to evolve spontaneously (and especially in a case of corneal surgery) form of ectasia. The choice of an optimal threshold led to a sensitivity of 93% and a specificity of 92% for the screening of Keratoconus fruste.

These results were responsible for the design of thealgorithm of the SCORE Analyzer (SCORE: 'Screening Corneal Objective Risk of Ectasia') based on the use of a combination of specular topographic indices of Placido and elevation with some power discriminant (existence of a statistically significant difference in the averages calculated between the groups tested for each of these indices). Among the clues with a statistically significant difference between the two groups, to have been selected Twelve of the most discriminating. Each of these indices was used and contributed to the calculation of a score, with a discriminant function that combined linearly.

The score allows you to sort comments according to the degree of similarity with likely corneas of spontaneous ectasia (evolution towards a more pronounced form of Keratoconus).  The threshold chosen, which corresponds to the border to separate the samples studied, with maximum specificity and sensitivity has been adjusted to take the value 'zero '.

A positive score (> 0) predicted a suspicious cornea, while a negative score (< 0) predicted a normal cornea. The value of the score is positive, more the topographical characteristics of the cornea under review are close to those of Keratoconus, and vice versa. The realization of corneal refractive surgery must be reconsidered in case of positive, even weakly negative value depending on the clinical context (strong myopic correction, atopic, etc.)

  Examples of SCORE software

 

Here is an example of map obtained from an Orbscan acquisition in a patient with a very asymmetrical form of Keratoconus (the left eye had a proven form).

Keratoconus subclinical: positive score

Right eye of a patient with an advanced form of the left eye Keratoconus. Inspection of the maps (Quadmap) allows to objectify the presence of a astigmatism reverse a bit spotty, and thick corneal Central average slightly below average (510 microns). The Radar map stresses the emphasis on the difference between the average central product and the thickness at the thinnest point value; This value is correlated with a gradient of high thinning (map "Nathan thinning rate"). The score is positive.

The next case is a case of ectasia iatrogenic post LASIK, where the SCORE software could be used retrospectively and measure a positive Score in preoperative.

Before LASIK pre-operative and post-operative maps after ectasia left eye

Map topographic preoperative Orbscan of the right and left eye (top), for compound myopic astigmatism correction. There is a responsible corneal toricite of a slightly irregular oblique astigmatism (SRAX). The posterior Central rear elevation is marked, the average central thickness is above average.
Despite the respect of a residual rear wall greater than 300 microns, a corneal ectasia has occurred at the level of the right eye, some months after achieving a bilateral LASIK (postoperative maps down)

The preoperative maps, which aroused no particular suspicion, were retrospectively used to establish a map with the SCORE Analyzer software.

The Score of the right eye is positive, indicating that the combination of topographical indices allowed to come to a conclusion that the subjective inspection could easily guess: the presence of a susceptibility for the corneal ectasia.

score right eye

SCORE software has been used retrospectively from the review data preoperative Orbscan for right eye reaches of post LASIK ectasia. He points out some pre-operative anomalies (map Radar), as well as a thinning higher towards the Center (Nathan thinning rate) rate. The Score has a positive value.

 

The score of the left eye (no complicated evolution) was slightly negative:

score left eye

The value of the eye adelphe score calculated from the Orbscan examination before LASIK is slightly negative. The radar map highlights the presence of asymmetry of vertical curvature, and a lower vertical shift of the thinnest point.

 

Interest of SCORE software

 

The interest of an automated system of topographic subclinical Keratoconus detection novice is multiple. It allows all first to combine a set of discriminating clues from the specular topography, elevation and thickness in a single digit (the score), for a more comprehensive assessment of each studied cornea. It's the first automated detection method that combines clues of topographies Placido, elevation and pachymetrique, and this association has proven its superiority to the separate use one or the other of the two technologies)9,11).

 

Conclusion

 

Corneal topography is a reference tool for the detection of Keratoconus subclinical forms. Specular Placido topography allows to track some of these forms (forms labeled as suspicious Placido) subclinical, but its sensitivity and specificity are not sufficient to abstain from the collection of data of the elevation and the total product (tomography).

All these results converge and strongly suggest that the surgeon concerned to increase the sensitivity of detection of subclinical Keratoconus beginner must pay attention to the review of the posterior side of the cornea and its thickness, which is evaluated more comprehensively by an examination of the distribution pachymetrique (tomography map or product). SCORE Analyzer is designed to achieve this goal.

 

VIDEO: introduction to the SCORE software analyze (Bordeaux, 24 may 2014)

 

see the article linked on a study to validate the SCORE Analyzer: https://www.gatinel.com/2013/10/saad-gatinel-d-retrospective-testing-of-the-score-for-the-detection-of-ectasia-susceptibility-casee-report-of-ectasia-7-years-after-lasik-int-j-kerat-ect-cor-dis20132273-78/

 

 

References

 

(1) Saad, Gatinel D. Topographic and tomographic properties of form fruste keratoconus corneas. Invest invest Vis Sci. 2010; 51 (11): 5546-55

(2) HR, Zhao SZ, L, Tan DT Lim Wei. Impact and characteristics of unilateral keratoconus classified on corneal topography. J Refract Surg. 2011; 27 (10): 745 - 51

(3) Rabinowitz YS, Nesburn AB, McDonnell PJ. Videokeratography of the fellow eye in unilateral keratoconus. Ophthalmology. 1993; 100 (2): 181-6.

(4) Gatinel D, Saad A. The challenges of the detection of subclinical keratoconus at its earliest stage. Int J Keratoco Corneal Ectatic say 2012; 1 (1): 36-43.

(5) Muftuoglu O, Ayar O, Ozulken K, Ozyol E, Akıncı A. Posterior corneal elevation and back difference corneal elevation in diagnosing form fruste keratoconus in the fellow eyes of unilateral keratoconus patients. J Cataract Refract Surg. 2013; 39 (9): 1348-57

(6) Smadja D, Touboul D, Cohen A, Doveh E, Santhiago MR Mello GR, Krueger RR, Colin J. Detection of subclinical keratoconus year using automated decision tree classification. Am J invest. 2013; 156 (2): 237-246

(7) Bühren J, Kook D, Yoon G, Kohnen T. Detection of subclinical keratoconus by using corneal anterior and posterior surface aberrations and space thickness profiles. Invest invest Vis Sci. 2010; 51 (7): 3424-32

(8) fonts BM, Ambrosio R Jr, Salomão M, Velarde GC, Nose W. Biomechanical and tomographic analysis of unilateral keratoconus. J Refract Surg. 2010; 26 (9): 677-81

(9) Saad A, Gatinel D. Validation of a new scoring system for the detection of early form of keratoconus. Int J Kerat Ect Horn say 2012,1 (2): 100-108

(10) Saad A, Gatinel D. Association of corneal indexes for the detection of ectasia-likely corneas. J Refract Surg. 2012; 28 (3): 166; author reply 166-7

(11) Saad A, Hodge C, Lawless M, Gatinel D. Retrospective testing of a new method for detecting ectasia-likely corneas. J Cataract Refract Surg. 2011; 37 (10): 1907-8

 

Leave a comment

You can ask questions or comment on this content: for this, use the "comments" form located below. The questions and comments of a general interest will be processed and published, and the information provided on the relevant pages should be clarified or supplemented.

Your e-mail address will not be published. Required fields are indicated with *