Periorbital Biometric Measurements using ImageJ Software: Standardisation of Technique and Assessment Of Intra- and Interobserver Variability.

PURPOSE: To assess the reliability and repeatability of periorbital biometric measurements using ImageJ software and to assess if the horizontal visible iris diameter (HVID) serves as a reliable scale for facial measurements.
METHODS: This study was a prospective, single-blind, comparative study. Two clinicians performed 12 periorbital measurements on 100 standardised face photographs. Each individual's HVID was determined by Orbscan IIz and used as a scale for measurements using ImageJ software. All measurements were repeated using the 'average' HVID of the study population as a measurement scale. Intraclass correlation coefficient (ICC) and Pearson product-moment coefficient were used as statistical tests to analyse the data.
RESULTS: The range of ICC for intra- and interobserver variability was 0.79-0.99 and 0.86-0.99, respectively. Test-retest reliability ranged from 0.66-1.0 to 0.77-0.98, respectively. When average HVID of the study population was used as scale, ICC ranged from 0.83 to 0.99, and the test-retest reliability ranged from 0.83 to 0.96 and the measurements correlated well with recordings done with individual Orbscan HVID measurements.
CONCLUSION: Periorbital biometric measurements using ImageJ software are reproducible and repeatable. Average HVID of the population as measured by Orbscan is a reliable scale for facial measurements.

INTRODUCTION

Accurate orbitofacial biometric measurements are of vital importance to the ophthalmic and facial plastic surgeon in the management, surgical planning and evaluation of outcomes.[12] Clinical biometric measurements are most commonly obtained using a millimetre ruler, with the examiner sitting at the eye level of the patient. Boboridis et al. have reported ‘modest and clinically acceptable’ inter- and intraobserver variability of eyelid measurements using this technique in patients with various clinical conditions.[3] Clinical measurement techniques, though simple and practical for day-to-day patient care, do have limitations such as observer variables, parallax, variable illumination and patient movement. Moreover, clinical measurements are limited only to linear measurements and more complex parameters such as area and volume cannot be measured. When performing clinical trials with utmost scientific rigor, a more objective and reproducible method is desirable.

In an attempt to improvise on the clinical measurement technique, several digital photographic techniques have applied facial and periorbital measurements. The photographic technique has several potential advantages. It provides standard illumination and working distance, allows for masking of observers, and permits rapid transfer of images between institutions, thus facilitating data storage and archiving.[1] The digital photographs can also be further analysed using various software programs.

ImageJ software is a reliable measurement tool available from the National Institute of Health, Bethesda, Maryland, USA (http://rsb.info.nih.gov/ij).[4] It has been successfully used as a measurement tool in various subspecialties of medicine.[5678] ImageJ has been used for periorbital and facial measurements.[910] Although ImageJ provides an objective measure of periorbital biometry, its inter- and intraobserver variability in the measurement of clinically significant periorbital parameters is not known. In this study, therefore, we aimed to determine the intra- and interobserver variability of periorbital biometric measurements when measured using ImageJ software.

Another difficulty while performing objective periorbital and facial measurements is the lack of a reliable measurement scale. Having a millimetre ruler within the frame of the photograph can provide measurement scale but is not always practical. Horizontal visible iris diameter when measured by a reliable instrument like Orbscan can provide a good measurement scale for facial measurements.[11] The human corneal diameter (horizontal visible iris diameter) has fairly stable dimensions for a given population and can be thought of as a naturally available measurement scale for all facial photographs. In this study, therefore, we also evaluated if the average corneal diameter of the population can be used as a reliable measurement scale.[12]

Procedure

This was a prospective, observational comparative study involving 100 consecutive normal controls who visited the Department of Ophthalmic Plastic Surgery at LV Prasad Eye Institute, Hyderabad, India.

Study objectives

The study had two objectives: First, to evaluate the intra- and interobserver variability of periorbital measurements as performed using ImageJ software. Second, to assess if the mean horizontal visible iris diameter (HVID) of the study population can be reliably used as a reference scale for these measurements.

Pilot study to determine sample size

Since there were no such studies reported in the literature, a pilot study was conducted to arrive at a sample size. For the pilot study, a standardised face photograph of one individual was taken after obtaining an informed consent. Five recordings of the HVID of the right eye were obtained, and an average was taken as the reference measurement. The photograph was uploaded in ImageJ software as a TIFF format image at 300 dots per inch (dpi) resolution. Margin reflex distance was measured 10 times by two investigators (MNN and RR). With this data, a sample size was calculated with a standard deviation of 0.16 mm to obtain an 80% power of the study at 95% confidence interval so as to have a precision of 0.1 mm in the measurements. A sample size of 100 individuals was arrived at.

Study protocol

We recruited 100 normal controls presenting to the Department of Ophthalmic Plastic Surgery, fulfilling the inclusion and exclusion criteria as shown in Table 1.

Table 1

Inclusion and exclusion criteria used to enrol individuals in the study

The study was approved by the Institute Review Board and Ethics Committee of LV Prasad Eye Institute, Hyderabad, India.

The purpose and the procedure of the study were explained to all the participants, and a written informed consent was obtained. Apart from complete ocular examination, each individual underwent following two evaluations: measurement of HVID and standardised face photographs.

Measurement of horizontal visible iris diameter

HVID was measured in the right eye of each individual using Orbscan (Bausch and Lomb Zyoptix Orbscan IIz Anterior Segment Analyzer). Five readings were obtained for the right eye of each individual [Figure 1b]. In case of poor fixation or increased blink and errors, the Orbscan reading was aborted and all the five readings were repeated. The average HVID of the right eye was then calculated and entered into patient data sheet.

Figure 1

Each photograph was imported in Adobe Photoshop 7, magnified to ×600 and the centre of both the pupils was manually marked (a). Horizontal visible iris diameter was measured in the right eye of each individual using Orbscan (b). Image imported into ImageJ software and horizontal visible iris diameter was marked with a line tool. This line was set at the horizontal visible iris diameter value for that patient (as measured by Orbscan in mm) and taken as the reference for biometric measurement. This step is called ‘setting the scale’ (c). Once the scale is set, the measurements were recorded (d)

Face photographs

Face photographs of 100 consecutive individuals were taken after explaining the purpose, the procedure and obtaining the informed consent for the study. Images were taken in a standardised manner with a Nikon D2X Digital SLR Camera with indirect lighting from a distance of 45 cm. All the images were uniformly cropped to include trichion to menton vertically and tragus to tragus horizontally. Image size was 6 inches (height) by 4.5 inches (width), stored at 300dpi in TIFF format.

Biometric measurements

A code number (between 1 and 100) was randomly assigned to the face photographs. Two physicians were asked to measure predefined periocular and facial biometric measurements on each photograph. Each physician repeated the whole set of measurements, to obtain a final set of two measurements.

Each photograph was imported in Adobe Photoshop 7, magnified to ×600 and the centre of both the pupils was manually marked using the ‘pencil tool’ set at 1-pixel size [Figure 1a]. This was performed to enable all measurements along the mid-pupillary plane. The photographs were then opened in ImageJ. In each photograph, HVID was then manually marked using the line tool. HVID value for that patient (as measured by Orbscan) was taken as the reference measurement for the respective photograph. This step is called ‘setting the scale’ [Figure 1c]. Once the scale is set, the measurements were recorded [Figure 1d]. The biometric measurements that were evaluated are given in Table 2.

Table 2

Periorbital biometrics that was measured and the definition used while measuring the same on photographs

All measurements were recorded on a sheet against the coded number of the photograph. Two investigators (MNN and RR) performed the measurements twice (measurement 1 and measurement 2) at two different times. The two observers were masked to each other’s readings. Another masked observer (KS) performed the analysis of the measurements.

Statistical analysis

Data were analysed by calculating intraclass correlation coefficient (ICC), Pearson product-moment correlation coefficient and Bland–Altman plots.

Outcome measures

The first component of the study involved biometric measurements based on individual HVID of each individual. Intraobserver variability for each measurement was calculated for each physician by comparing measurement 1 to measurement 2. Interobserver variability for each measurement was calculated by comparing the average of each measurement of one physician to the other.

The second component of the study involved biometric measurements based on the ‘mean’ HVID of all individuals. All biometric measurements were repeated using the mean horizontal corneal diameter (11.34 mm). The intra- and interobserver variability was again calculated in a similar manner.

RESULTS

A total of 111 individuals fulfilling the inclusion criteria were enrolled in the study. Eleven photographs were excluded due to the poor quality, which could have precluded accurate analysis. The remaining 100 face photographs were analysed by two observers (MNN and RR) in a masked manner. The patient demographics and the HVID values as pleasured by Orbscan are given in Table 3.

Table 3

Demographics of 100 individuals who participated in the study

For measurements representing part I of the study (each patient’s own HVID reading was utilised for measurement), the mean and the standard deviation of each parameter (two measurements; both intra- and interobserver) is depicted in Table 4. The ICC values for the intra- and interobserver variability are depicted in Table 5. The test-retest reliability was measured using Pearson product-moment correlation coefficient and is depicted in Table 6.

Table 4

The mean readings and P values for intra- and interobserver measurements of various biometric parameters

Table 5

Intraclass correlation coefficient values for intra- and interobserver variability

Table 6

Test-retest reliability values for all biometric parameters

The intraobserver test-retest reliability varied from 66 to 100. The value was least for medial brow height (0.66) and highest for canthus to commissure (1.00). All the other parameters ranged above 0.90, indicating good reliability.

The interobserver test-retest reliability varied from 0.77 to 0.97. The value was least for medial brow height (0.77), and highest for canthus to commissure distance on the left (0.98), indicating good reliability. Bland–Altman scatter plots were generated for each measurement to look for the agreement between the two observers [Table 7].

Table 7

Scatter plot for Bland–Altman analysis for data of margin reflex distance-1 parameter

The differences (investigator-1 minus investigator-2) are plotted on Y-axis and the average value of the two investigators is plotted on X-axis. Mean difference (bias) is depicted along the central line and 2 standard deviations are shown above and below the mean bias line. These values (mean + 2 standard deviation) are known as ‘limits of agreement’. The plots depicted good agreement between the two observers in all the measurements.

The ICC values for the intraobserver variability for the measurements using the ‘mean’ HVID of 100 individuals (11.34 mm) as the scale ranged from 0.75 to 0.99 and the test-retest reliability for the same ranged from 0.62 to 0.94 [Table 8].

Table 8

Intersession intraclass correlation coefficient and test-retest reliability of different measurements by using the ‘mean’ horizontal visible iris diameter as a scale

DISCUSSION

The inter- and intraobserver variability in the measurement of clinically significant periorbital parameters are not known. This study aimed to determine the intra- and interobserver variability in the periorbital biometric measurements performed using ImageJ software.

ICC values were >90 for both intra- and interobserver measurements and ranged from 0.75 to 0.99 for measurements using the ‘mean’ HVID value. The test-retest reliability for the intra- and interobserver measurements was good (>90%). For the measurements using mean HVID, it ranged from 0.62 to 0.97. This indicates that the periorbital biometric measurements using ImageJ software are reproducible and repeatable. When the same measurements were repeated using the mean HVID of the study population (11.34 mm), the measurements were still comparable to the previous measurements. This indicates that the ‘mean’ HVID of the population (as measured by Orbscan) is a reliable scale for facial measurements. The latter inference has the potential to allow reliable facial measurements if the mean HVID of the population is known.

The ICC and test-retest reliability values were low for brow height as compared to the other measurements. The arrangement of the eyebrow hair is sparse, and less distinct a landmark, leading to variability. Further, in some individuals, eyebrow hair may not extend more medially, so as to be in line with the medial canthus, which was our measurement reference. Measurement of the lateral canthus was also difficult in some cases, due to lateral lash ptosis that precluded accurate identification of the lateral canthus.

Our study for the first time outlines a standardised approach to periorbital measurement using ImageJ software and HVID. In conclusion, our study found that reliable periorbital biometric measurements can be obtained using ImageJ software. The intra- and interobserver variability is low. We also found that the mean HVID of the population can be a reliable scale, against which other facial measurements can be made.

Financial support and sponsorship

Nil.

Conflicts of interest

None.