Grey value contrast sensitivity of dental practitioners in function of luminance and age

Abstract

Objectives:

Grey value perception is important in viewing and interpreting X-ray images. It is possible that ageing decreases the number of gray values that a person can distinguish. This hypothesis was tested in a group of dental practitioners from students to experienced dentists using a 12-bit grey scale.

Methods and materials:

A custom-made computer software was used based on the principle of the “just noticeable difference” (JND). Observers were shown a picture of an outer square with increasing grey value in which a smaller square of different grey value popped up in changing positions. As soon as the observers could see the difference, they clicked the inner square. The grey values were shown with a 12-bit depth in batches of 400 grey value steps. 59 dentists of 3 age groups (20–24, 25–45 and 46–70 years) participated in the study. A subgroup of 20 practitioners performed a validation test with test-retest and test in reversed grey value sequence (white to black). JND was calculated and plotted against grey value using a third-order polynomial function. These curves were compared statistically.

Results:

Test-retest, also in reversed order did not yield different curves (p values between 0.79 and 0.97). Curves between different age groups showed significant differences, with older practitioners needing more contrast to accomplish the task. Contrast sensitivity showed an optimum in the darker third of the grey scale (values around 1200).

Conclusion:

Age and grey value played a significant role in grey value perception by dental practitioners.

Introduction

In the digital era medical images are datasets, recordings or reconstructions of physical properties of the area of interest, the field of view, most often by grey value levels. Image acquisition devices are performant and the information captured is stored with an increasing number of grey value levels expressed as bit depth. Although given its technological potential, computer-aided image capturing and processing is extremely performant, it depends on the operator to inspect the images, draw adequate conclusions and come to a diagnosis or treatment decision. Compared to recent technological advances, like the increased resolution or grey level depth, the human eye has its limitations. How much grey levels can be displayed on a specific monitor and how are these grey levels to be used?¹ For a human to explore and detect all information in an image it must be represented in optimal conditions. How can the grey level intensities available on a display be used as a function of the grey value levels provided by the image capturing process to allow for optimal diagnosis?² The DICOM standard incorporates a transformation of raw information—images—to alter the differences in grey values into just noticeable differences (JND) when displayed on calibrated high end medical displays in an optimized viewing environment.^3–5 JND is defined as the difference in displayed intensities that can be detected by an average operator in 50% of cases. The idea is that to allow interpretation of a difference in grey value levels in a digital image, the user must be able to see the difference in intensities used to display this difference. The strategy to use a high end calibrated screen, a standardized and optimized environment and a transformation of the raw data into grey level differences that are just noticeable presumes that the transformation is optimal for all users or all users have “average” eye sight over the whole grey value range.

Most dental imaging is done based on the measurement of radio-opacity measured through the attenuation an X-ray beam by the material placed between X-ray source and capturing device. In a majority of cases these images are interpreted on the spot using displays at the dental chair or in the dental office. The viewing environment is multi purpose and not optimized for image viewing and interpretation. Moreover, unlike in radiology departments of hospitals, the displays are often “Commercial Of The Shelf” (COTS). The prerequisite for a transformation optimal for all dental practitioners in combination with the type of display and the viewing environments is highly unlikely. When focusing on the practitioner there are huge age differences between starting and older practitioners. Illness such as diabetes can influence vision, but even without illness eye sight is influenced by fatigue.^6–8

Much research effort has been dedicated to evaluate and compare the imaging technology (captors, processing display), for instance for the detection of dental caries, endodontic pathologies or periodontal diseases.^9–11 However, being one of the main links in the diagnostic chain, the observer has received little attention. Many studies were performed using a comparatively small number of observers (2 to 4) and potentially, a large unrecognized source of variation (leading to a loss of power) may have originated in the observer’s different visual contrast sensitivity.

Grey value perception has been explored in several studies.^12–14 In order to estimate the JND in a standardized way the experimental methodology described by Okkalides was modified to be used for a 12-bit grey value display.^12,15 Through this experiment patterns of grey level sensitivity are explored for several dental practitioners and the effect of age on eye sight.

Methods and Materials

59 dental practitioners participated in the study between the age of 20 and 70. They were shown a picture of an outer square and a smaller square of different grey value. Every 2.5 s the smaller square was randomly repositioned in a regular three by three grid with a random increase in grey value from the outer square of between 1 and 50 grey value levels. The same sequence of grey value/location combinations was used for all observers. As soon as the observers could see the difference, they were asked to point at the smaller square (see Figure 1). The grey values were shown with a 12 bit depth (4096 grey values). The outer square measured 7.5 cm by 7.5 cm the smaller square 1.5 cm by 1.5 cm.

Figure 1.

Screenshot showing the outer 7.5 cm by 7.5 cm square in which every 2.5 s a 1.5 cm by 1.5 cm smaller square of different grey value popped up randomly located on a regular three by three grid up with a random increase in grey value from the outer square of between 1 and 50 grey value levels (a). On the left image (b), a frame indicates this popping up square on the second row and third column of the grid.¹⁵

We opted for a gradual increase of background grey value from a dark background to a light background to ensure that blinding through sudden changes in background would not introduce carry over effects. Drawback of this approach is that concentration and fatigue can play a role in the grey value sensibility patterns. To assess reliability, the possible influence of concentration and fatigue, 20 practitioners were asked to perform the experiment four times. The first two repetitions with an intermediate period of at least one week up to one month. The second and the third testing followed each other immediately. The fourth and final test was performed on a different day with background grey value levels gradually decreasing from light to dark. The four repetitions were finally completed by 19 practitioners, one practitioner only performing the first test.

The representation software was custom made using MatLab R2018a version (9.4.0.813654) (The MathWorks, Inc., Natick, MA). Statistical analysis was performed using Matlab and GraphPad Prism 6.05 (GraphPad Software, La Jolla, CA). One commercial of the shelf Dell Latitude E5540 laptop was used for all experiments. The scoring was performed in a dark room without windows or ambient light.

Data were then retrieved from the computer and contrast sensitivity was calculated for each background grey value bin of 400 width. JND, expressed as the difference in grey value with a probability of 50% to be detected was estimated using logistic regression with grey value difference as predictor for consecutive background grey value bins. Maximum likelihood estimation was used.¹⁵ The resulting JND estimates were plotted per 400 grey value bins. Using GraphPad Prism (GraphPad, La Jolla, CA), a third-order polynomial was fitted per test run. Mean polynomials were compared by applying the Akaike information criterion. Using this method, the model of a common regression line was tested against the probability of separate regression lines for different data sets. This test was performed to compare the forward/reverse test runs, the several repeated forward test runs and eventually the three different age groups.

Results

The validation tests showed that performing the test run “forward” (increasing background values) could be described with the same regression curve similar to the JND curve with “reverse” (decreasing background values) with a probability of 96% (see Figure 2a). In Figure 2b JND as a function of the background intensity for test, retest and the immediate second retest is presented. The retest outperforms the first test, while the immediate second retest has the on average the worst results, consistently over the whole background grey value range.

Figure 2.

(a) JND values for test with increasing background intensity (forward) and decreasing background intensity (reversed). The line indicates the common regression line for both test runs. (b). JND with increase in background light intensity for test, retest and immediate second retest. The regression lines are given as common curve for test-retest1 (continuous line), retest 1 (dashed line) and retest 2 (dotted line).

The curves representing JND as a function of background grey-value corresponding to the separate age classes differed significantly from each other (probability 98% AIC difference 13.6%, Figure 3). The r² of the regression lines was situated around 0.4. Not surprisingly the JND for the younger age class was systematically lower than the middle age class and the higher age class. The middle age class started with a JND comparable to the younger age class to end with a JND equal to the older age class. The JND curve for the older age class was consistently higher than the JND curve for the other age classes except for the lightest background value. Observe that the variability for the highest classes there was a substantial increase in variability. The variability at start was most likely due to adaptation to the test at start, despite the relatively long periods between presented images (2.5 s).

Figure 3.

JND as a function of background intensity for the age classes 20–24 years, 25–45 years and 46–70 years. There was a significant difference between the three regression lines (p = 0.002).

Although on average there is an increase in JND as the background lighting increases this tendency is not applicable in general. Different JND profiles exist. in Figure 4 three cases are presented, the first case shows a more or less linear increase of the JND with increasing background lighting, the second profile is flat, the JND remains constant over the whole range of background intensities, and the last profile shows a general JND increase with a decrease at the highest background intensity.

Figure 4.

Three examples of different grey value distinction patterns. Light grey stars indicate a noticed difference, dark grey stars indicate failure to detect. Circles indicate the estimated JND. (a) Fairly monotonous linear increase in JND with increase in background light intensity. (b) Overall low and fairly constant JND values. (c) Adaptation problems at startup followed by a steady decline and a decrease at the highest light intensity.

Discussion

This study focusses on the dental practitioner and one relevant aspect, age, controlling for environmental factors by means of a fairly simple test setting. This approach has undeniable practical advantages over pre-existing approaches. Haak et al used an adaption of the test pattern of the Society of Motion Pictures and Television Engineers (SMPTE) as a basis.¹³ This background is not neutral and might cause bias. Other studies use a limited number of test images to determine the JND e.g. Bender et al used seven medical images and presented them to twenty radiologists for evaluation.¹⁶ The aim of the study was not the individual radiologists viewing capacity, but to evaluate the image bit depth preference given a high end monochromatic display. In order to assess possible differences in JND between Chinese and Dutch, Qin et al used seven images to conclude that there are differences in black level JND but that these differences might be influenced by the fact that the “viewers looked to a different part in the images rather than to an ethnic difference”.¹⁴ It is clear that images with a neutral content are to be preferred.

In the original experiment by Okkalides one of the squares of a four by four grid of squares was randomly selected and increased by one grey value level with respect to the 15 remaining squares.¹² In our test the increase in grey value levels was chosen between one and 50 to accommodate for the increased grey value depth of the present medical images and the increased display quality since Okkalides conducted his experiments (based on eight bit depth, allowing for 264 steps). In contrast to other studies the study by Okkalides explored all available background grey values and all 16 positions, clearly this can lead to fatigue of the observer. The random element in the conducted experiment allowed to reduce the number of viewings by the observers and allowed for a statistical approach to the JND estimation. The results of our validation tests, including repetition and reversal of light intensity suggest that the attempt was sufficiently successful for conclusions on group level.

In this study we showed that there were considerable interindividual differences in grey value perception with age as an influencing factor. These differences were significant but not tremendous. This was in line with previous experimental data demonstrating a decrease of contrast sensitivity with age.^17,18 However, this decrease is slow in the age group between 55–65 and only clearly increased in an age group in which generally, dental practitioners cease their professional activity.¹⁹

As JND values were systematically higher in the retest run (performed immediately after the first retest, Figure 2), operator fatigue could have played a role.²⁰

Independently of age, contrast sensitivity was lowest in the high luminance range (light grey and white). This might, next to other factors as lesion size, explain the lower detection rate of enamel caries lesions. Intuitively, this would call for reversing the grey scale for these lesions although in an experiment this could not be substantiated.²¹ However, this study found that reversed images showing dentinal lesions decreased their chance of detection, especially at low resolution. Possibly, these lesions were displayed in an unfavorable part of the grey scale. Li et al (2002) showed that transforming radiographic images so that possible carious lesions were displayed at an ‘optimal’ grey scale value could improve caries detection.²² In their study, the optimum was set at about half-way of the scale, however, as seen in Figures 2 and 3, about 1/3 of the scale in the lower luminance range seems to be preferable.

The DICOM standard, in combination with the use of calibrated monitors and controlled environments compensates on average for the perception of the human being.⁴ However, we showed that there is a considerable individual variability in viewing capabilities between individual dental practitioners. These variations could be classified into different profiles. Therefore, a transformation optimal for one person will possibly decrease the perception for another person. The overall substantial variability in JND indicates that a one-fits-all transformation for patterns that conform the general tendency will most often be sub optimal. Further research will be performed to assess the effect on clinical diagnosis. In the meantime, practitioners using uncalibrated COTS monitors should use the brightness function on the display to change the grey-scale of a specific area of interest in order to roughly obtain the optimal luminance to allow task-specific image observation.²³

Conclusions

Gray value perception is individual, not all humans are alike with respect to their visual capacities. Grey value contrast sensitivity deteriorates with age, however all age groups have an inferior contrast sensitivity in the high luminance (white) range. Individual transformations are needed to optimize the gray value perception for everybody. Until this is technologically possible, a calibration of commercial screens according to the DICOM norm may already bring some improvement.