Original Investigation
The Efficacy of a Novel Mobile Phone Application for Goldmann Ptosis Visual Field Interpretation Robi N. Maamari, B.S.*, Michael V. D’Ambrosio, Ph.D.†, Jeffrey M. Joseph, M.D.*, and Jeremiah P. Tao, M.D., F.A.C.S.* *Gavin Herbert Eye Institute, University of California, Irvine; and †Department of Bioengineering, University of California, Berkeley, California, U.S.A.
Purpose: To evaluate the efficacy of a novel mobile phone application that calculates superior visual field defects on Goldmann visual field charts. Methods: Experimental study in which the mobile phone application and 14 oculoplastic surgeons interpreted the superior visual field defect in 10 Goldmann charts. Percent error of the mobile phone application and the oculoplastic surgeons’ estimates were calculated compared with computer software computation of the actual defects. Precision and time efficiency of the application were evaluated by processing the same Goldmann visual field chart 10 repeated times. Results: The mobile phone application was associated with a mean percent error of 1.98% (95% confidence interval[CI], 0.87%–3.10%) in superior visual field defect calculation. The average mean percent error of the oculoplastic surgeons’ visual estimates was 19.75% (95% CI, 14.39%–25.11%). Oculoplastic surgeons, on average, underestimated the defect in all 10 Goldmann charts. There was high interobserver variance among oculoplastic surgeons. The percent error of the 10 repeated measurements on a single chart was 0.93% (95% CI, 0.40%–1.46%). The average time to process 1 chart was 12.9 seconds (95% CI, 10.9–15.0 seconds). Conclusions: The mobile phone application was highly accurate, precise, and time-efficient in calculating the percent superior visual field defect using Goldmann charts. Oculoplastic surgeon visual interpretations were highly inaccurate, highly variable, and usually underestimated the field vision loss. (Ophthal Plast Reconstr Surg 2014;30:141–145)
V
isual field (VF) testing is commonly included in the preoperative evaluation for functional upper eyelid lifting surgery (e.g., ptosis repair, blepharoplasty, brow lift).1 The specific aim of this test is to quantify the amount of superior visual field (SVF) obstruction by comparing results with the eyelids taped or manually elevated to results with the eyelids in the baseline position. The Center for Medicare Services and other thirdparty payers have defined a 30% loss in the overall SVF as a threshold for medical necessity.2,3 Accepted for publication August 30, 2013. Presented at the American Society of Ophthalmic Plastic and Reconstructive Surgery Fall Scientific Symposium, November 2013, New Orleans, Louisiana. Supported, in part, by an institutional Research to Prevent Blindness grant. The authors have no financial or conflicts of interest to disclose. Address correspondence and reprint requests to Robi N. Maamari, Department of Ophthalmology, Gavin Herbert Eye Institute, University of California, Irvine, CA. E-mail: [email protected] DOI: 10.1097/IOP.0000000000000030
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
While straightforward on static, automated (e.g., Humphrey) VF tests that provide a numeric score of number of stimuli seen by the patient and number of stimuli presented, the percent SVF defect on a kinetic manual (Goldmann) ptosis VF test may be problematic because calculations require one to determine area under curvilinear, noncircular, and often irregular curves. One commonly used method to quantify the overall Goldmann SVF defect is to visually estimate the amount of change in the taped and untaped VF. Meyer et al.4 and Riemann et al.5 described using the height of the SVF at the 90° meridian to determine percent SVF impairment; however, this calculation is predictably inaccurate because the area in question is not rectangular, nor perfectly circular, as the simple mathematical formula assumes. Moreover, many patients have disproportionate superonasal or superotemporal SVF loss. The authors are unaware of any standardized tool, software, or device that calculates the percent VF obstruction on Goldmann VF testing. Mobile smart phone and portable tablet devices are becoming more integrated in medical practice and may be leveraged as a tool for these SVF defect calculations.6 This study introduces a novel mobile phone application that calculates percent SVF defects using a Goldmann VF test and evaluates its efficacy compared with oculoplastic surgeons’ visual estimates and the actual computer-calculated values.
METHODS Questionnaires. Fourteen practicing ophthalmology board certified and fellowship-trained oculoplastic surgeons who routinely use Goldmann VF testing in their practice to evaluate for ptosis repair completed a web-based questionnaire that comprises 10 representative Goldmann ptosis SVF, each with 2 drawn curves: a lower curve representing the patient SVF at baseline (untaped eyelid); and an upper curve representing the patient SVF with the eyelid elevated (usually taped). The baseline SVF area was defined as the area delineated by the untaped VF curve and the horizontal axis (0° and 180° meridian). The unobstructed SVF was defined as the area delineated by the taped VF curve and the horizontal axis. Oculoplastic surgeons were instructed to use their method in practice to visually estimate the percentage of SVF defect, defined as the difference between the unobstructed and baseline SVF areas divided by the unobstructed SVF area. Mobile Phone Application. An iPhone (Apple Inc., Cupertino, CA, U.S.A.) application (app) was created to calculate SVF defect percentages. The app uses an iPhone 4S camera to capture images of the Goldmann SVF chart. Morphological image processing techniques detected the taped and untaped curves from the captured image. Color threshold differences were used to detect and distinguish the 2 curves, so the untaped and taped curves were drawn using 2 different colors (red, blue, or
141
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
R. N. Maamari et al.
FIG. 1. A, Screen shot of the application in preview mode. The user aligns the cross target with the vertical and horizontal axes and the outer circle to equally encompass the circle of eccentricity. Chart 1 is shown. B, Result screen with calculation of SVF defect. Once the chart is imaged, the application performs an automated calculation of the SVF defect and displays an animated representation of the SVF curves. C, Representative Goldmann visual field of Chart 2. D, Representative Goldmann visual field of Chart 3. green). A reference target was displayed on the camera preview screen that the user was instructed to align with the vertical and horizontal axes of the Goldmann charts (Fig. 1A). Tapping on the touchscreen captured images, and an automated algorithm calculated and displayed the SVF defect percentage and a representative image of the chart (Fig. 1B). The app was tested using the same 10 Goldmann SVF charts in the questionnaire given to the oculoplastic surgeons (Fig. 1C,D). To assess the precision and time efficiency of the app, 1 author (R.M.) imaged 1 chart 10 times. The mean SVF defect measurement and the mean time to obtain results were calculated with 95% CIs. The time to obtain a result was defined as the time elapsed between the start of the app and the display of the results screen. Comparison of Superior Visual Field Defect Calculations. Actual SVF defects were calculated using the open-source ImageJ software (National Institute of Health, MD, U.S.A.). After scanning the Goldmann charts to a computer, ImageJ software was used to manually create
142
geometric shapes from the taped and untaped SVF curves to calculate the areas within, from which the percent defect was then determined. The percent error (defined as the absolute value of the difference between the tested method area and the actual area divided by the actual area) was calculated for all 10 charts for the graders’ estimates and the iPhone app. The mean percent errors with 95% CIs were calculated for the graders’ estimates and the iPhone app calculations.
RESULTS Table 1 shows the responses provided by the oculoplastic surgeons, the measurements calculated by the iPhone app, and the calculated percent errors. Questionnaire. The average of the mean percent error of the oculoplastic surgeons’ visual estimates of SVF defects was 19.75% (95% CI [CI], 14.39–25.11%) and the standard deviation (SD) of the graders’ mean
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
(0.93)
(0.31)
(0.66)
(0.65)
(0.30)
(0.38)
(0.53)
(0.97)
Chart 1
Chart 2
Chart 3
Chart 4
Chart 5
Chart 6
Chart 7
Chart 8
Chart 9
1.98 1.56
App 0.19 (0.59) 1.55 (0.92) 0.87 (0.31) 0.48 (0.66) 0.59 (0.65) 3.99 (0.29) 3.81 (0.37) 2.52 (0.54) 4.26 (0.93) 1.57 (0.59)
19.65 19.76
A 32.39 (0.40) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 7.69 (0.60) 66.53 (0.10) 4.70 (0.40) 4.79 (0.50) 7.59 (0.90) 14.36 (0.50) 15.16 15.03
B 32.39 (0.40) 14.06 (0.80) 3.13 (0.30) 6.18 (0.70) 0.01 (0.65) 39.75 (0.18) 21.48 (0.30) 31.45 (0.36) 0.40 (0.97) 2.77 (0.60) 15.57 15.95
C 1.42 (0.60) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 49.79 (0.15) 4.70 (0.40) 4.79 (0.50) 10.67 (0.87) 2.77 (0.60) 19.71 14.12
D 32.39 (0.40) 24.80 (0.70) 32.19 (0.21) 12.02 (0.58) 19.99 (0.52) 43.09 (0.17) 0.54 (0.38) 6.70 (0.49) 2.46 (0.95) 22.93 (0.45) 25.14 18.61
E 32.39 (0.40) 24.80 (0.70) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 66.53 (0.10) 21.48 (0.30) 4.79 (0.50) 2.46 (0.95) 31.49 (0.40) 46.39 17.34
F 57.74 (0.25) 30.17 (0.65) 67.71 (0.10) 46.91 (0.35) 38.48 (0.40) 66.53 (0.10) 60.74 (0.15) 42.88 (0.30) 12.72 (0.85) 40.05 (0.35) 25.43 18.15
G 32.39 (0.40) 19.43 (0.75) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 66.53 (0.10) 4.70 (0.40) 33.29 (0.70) 7.59 (0.90) 22.93 (0.45) 17.48 14.67
H 32.39 (0.40) 14.06 (0.80) 19.28 (0.25) 8.99 (0.60) 15.38 (0.55) 49.79 (0.15) 4.70 (0.40) 4.79 (0.50) 2.46 (0.95) 22.93 (0.45)
Graders’ estimates
9.63 11.09
I 1.42 (0.60) 3.32 (0.90) 19.28 (0.25) 0.12 (0.66) 7.69 (0.60) 33.05 (0.20) 21.48 (0.30) 4.73 (0.55) 2.46 (0.95) 2.77 (0.60) 14.94 12.51
J 32.39 (0.40) 24.80 (0.70) 35.42 (0.20) 16.57 (0.55) 7.69 (0.60) 16.31 (0.25) 0.54 (0.38) 10.51 (0.47) 2.46 (0.95) 2.77 (0.60) 7.73 4.59
K 7.03 (0.55) 3.32 (0.90) 3.13 (0.30) 6.18 (0.70) 7.69 (0.60) 16.31 (0.25) 12.55 (0.43) 10.44 (0.58) 1.43 (0.96) 9.22 (0.53) 18.06 16.62
L 32.39 (0.40) 3.32 (0.90) 35.42 (0.20) 1.40 (0.65) 23.07 (0.50) 49.79 (0.15) 13.62 (0.33) 4.79 (0.50) 2.46 (0.95) 14.36 (0.50)
24.76 15.49
M 23.93 (0.45) 19.43 (0.75) 35.42 (0.20) 1.40 (0.65) 30.76 (0.45) 56.48 (0.13) 21.48 (0.30) 33.35 (0.35) 7.59 (0.90) 17.79 (0.48)
16.83 13.32
N 1.42 (0.60) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 16.31 (0.25) 21.48 (0.30) 4.79 (0.50) 2.68 (1.00) 40.05 (0.35)
(0.49)
(0.93)
(0.49)
(0.34)
(0.16)
(0.53)
(0.60)
(0.22)
(0.78)
Mean estimates (0.45)
Actual, iPhone app, graders’ estimates. Bolded values, percent error values calculated based on actual SVF defect (reported as percentages); Oculoplastic surgeons’ average mean percent error and standard deviation, 19.75% and 9.29%, respectively; Values in parentheses, visual field defects reported as fractions with 1.00 defined as a 100% defect.
Mean percent error Mean SD
Chart 10 (0.58)
Actual SVF defect (0.59)
TABLE 1. Superior visual field defects
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
Mobile Phone App for Goldmann Visual Field Defect
143
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
R. N. Maamari et al.
TABLE 2. Mobile phone app measurements using single chart
Mean SD
Time
SVF defect
Percent error
15.3 13 15.8 9.6 13.6 11.5 9.1 9.5 14.7 17.2
58.71 58.84 59.04 58.43 59.07 58.14 59.02 58.84 57.82 58.2
0.76 0.55 0.2 1.24 0.16 1.73 0.23 0.54 2.26 1.62
12.9 2.9
58.61 0.44
0.93 0.74
SVF, superior visual field; SD, standard deviation. Time is reported in seconds. SVF defect and percent error are reported as percentages.
percent errors was 9.29%. The mean percent error of the oculoplastic surgeon estimates ranged from 7.73% to 46.39%. The SD of the individual grader estimates ranged from 4.59% to 19.76% (Table 1). Mobile Phone Application. The mean percent error of the iPhone application was 1.98% (95% CI, 0.87%–3.10%) with a SD of 1.56% (Table 1). When comparing 10 measurements taken with the iPhone app using a single chart, the average percent error of the 10 measurements was 0.93% (95% CI, 0.40%–1.46%) with a SD of 0.73%. The percent error of the 10 app measurements of the same chart ranged from 0.16% to 2.26%. The average time for analysis of a single Goldmann SVF chart using the app was 12.9 seconds (95% CI, 10.9–15.0 seconds) with a SD of 2.9 seconds (Table 2).
DISCUSSION SVF testing is a common step in the preoperative evaluation for functional upper eyelid surgery. According to an American Society of Ophthalmic Plastic and Reconstructive Surgery national survey on current ptosis management, 87.4% of participants reported using some form of VF preoperatively.1 In addition, many third-party insurers require formal VF quantification of the field defect in their coverage determination for functional eyelid lifting surgery. Riemann et al.5 reported that Goldmann manual kinetic and Humphrey automated static testing are the most common tests. They found both to be effective in documenting ptosisassociated VF loss. However, they concluded that Goldmann testing was more sensitive and 5 times faster than the Humphrey test. Alniemi et al.7 demonstrated that Goldmann testing remained significantly more time-efficient even compared with current Humphrey software. In addition, they reported that 7 of 10 patients preferred Goldmann over Humphrey testing. Though many clinicians prefer Humphrey automated static testing for ease and objectivity in the general ophthalmology setting, Goldmann perimetry remains a workhorse for ptosis and other VF analysis.7,8 A limitation of Goldmann ptosis VF tests is that the interpretation may be more subjective than automated tests. The included cohort of fellowship-trained oculoplastic surgeons had an average of 19.75% error from the actual SVF defect on Goldmann. Oculoplastic surgeons, on average, underestimated the SVF defect, with 8 of the 10 charts showing a statistically significant underestimation (p < 0.05) (Fig. 2). These findings
144
FIG. 2. Graph demonstrating the mean oculoplastic surgeon estimate of percent superior visual field defect and the computer software calculated (actual) value for the 10 Goldmann charts. Error bars represent the 95% confidence interval. SVF, superior visual field.
suggest that current standards of VF interpretation may underdiagnose functional eyelid ptosis or dermatochalasis, at least by present VF defect standards. The reasons for underestimation are not certain, but the authors hypothesize that human visual estimation bias might play a role. First, figures of the same shape but different sizes are perceived to vary less in size than the actual size change.9 The similarly shaped untaped and taped curves may contribute to this underestimation bias. Second, visual estimations comparing areas of 2 shapes are often made using a dominant single dimension measurement to compare the 2 areas.10 In the Goldmann charts, the 90° superior meridian provides a convenient dominant dimension to compare the areas of the taped and uptaped areas. However, employing Chart 2 as an example (Fig. 1C), utilizing the superior meridian as a dominant dimension (10° untaped and 40° taped), leads to an underestimation of the SVF defect (75% calculating area as a rectangle using the superior meridian as the height as compared with 93.75% when calculating the area as a circle using the superior meridian as the radius). Interestingly, the mean oculoplastic surgeon estimate for this chart was 78%, which approaches the 75% estimate using the superior meridian for the rectangular area, and the actual defect was calculated as 93%, which is better approximated by the circular area estimate. Last, oculoplastic surgeons may have a subconscious bias to be conservative and underestimate the SVF defects to avoid third-party payer medical necessity scrutiny. Further research is necessary to clarify the reasons for underestimation by visual interpretation methods. The app may address the variability in visual SVF defect estimates and the lack of a standardized method for Goldmann test analysis. The iPhone app was shown to be statistically superior to the oculoplastic surgeons’ estimates for the calculation of the SVF defect (p < 0.05). This application accurately calculates the overall SVF defect, whereas other methods only provide estimates. For instance, as Anderson et al.11 described, using the height of the curves at the 90° meridian is inaccurate for most Goldmann ptosis field tests but especially when the curve is irregular or unbalanced such as in lateral skin overhang or lateral ptosis with more obstruction temporally.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
With mobile smartphone and portable tablet devices becoming more integrated in medical practice, the percent field defectcalculating app may be easily integrated in practice. However, further refinements may improve the usefulness of the app. Some limitations of the study include the relatively small number of oculoplastic surgeons surveyed. Although the app is highly intuitive and user-friendly, there may be a small learning curve. However, an underestimation bias does not appear to affect this study as the times and the percent errors recorded in Table 2 did not show evidence of improvement in the chronologically listed measurements. The accuracy and precision of the app should, in theory, be equivalent to the computer software values. However, since the app requires the user to align the target to the Goldmann chart, misalignment and varying degrees of camera tilt result in some variance that forestalls perfect precision and accuracy. Nevertheless, Table 3 demonstrates that the 10 measurements of the same chart had only a 0.93% error even with the variations in tilt and misalignment during each individual measurement. Another limitation of the app is the required colorcoded VF curves. Future versions will attempt to eliminate this requirement to increase the ease-of-use of the app. One technological limitation of this study is that the app was only designed to run on the iPhone operating system. Versions for other widely used mobile phone operating systems, such as Android (Google, Inc., Mountain View, CA, U.S.A.), are in the next phases of development. Development of the app for other operating system will be rapidly completed, as much of the app code can be used across various platforms. Further pilot testing may be necessary to ensure smooth and robust functionality. At the completion of pilot testing, the app will be released as a free iPhone app. In summary, this study demonstrates that current methods for SVF defect estimation on ptosis Goldmann VFs may be highly variable and underestimate the actual amount of ptosis. This study also introduces a mobile phone application that can rapidly calculate both precise and accurate SVF defect measurements.
Mobile Phone App for Goldmann Visual Field Defect
REFERENCES 1. Aakalu VK, Setabutr P. Current ptosis management: a national survey of ASOPRS members. Ophthal Plast Reconstr Surg 2011;27:270–6. 2. Medicare Local Coverage Determination for Blepharoplasty, Blepharoptosis, and Brow Lift (Effective 8/18/2008–5/1/2013). Available at: http://www.cms.gov/medicare-coverage-database/details/lcd-details.aspx?LCDId=27474&ContrId=161&ver=59&Con trVer=2&Date=08%2f26%2f2013&DocID=L27474&bc=AAAAA AgAIAAAAA%3d%3d&. Accessed June 27, 2013. 3. United Heath Care Services, Inc. Blepharoplasty, Blepharoptosis, and Brow Ptosis Repair Coverage Determination Guidelines (Effective 4/1/2013). Available at: https://www.unitedhealthcareonline.com/ ccmcontent/ProviderII/UHC/en-US/Assets/ProviderStaticFiles/ ProviderStaticFilesPdf/Tools%20and%20Resources/Policies%20 and%20Protocols/Medical%20Policies/Medical%20Policies/ Blepharoplasty_CD.pdf. Acessed June 27, 2013. 4. Meyer DR, Linberg JV, Powell SR, et al. Quantitating the superior visual field loss associated with ptosis. Arch Ophthalmol 1989;107:840–3. 5. Riemann CD, Hanson S, Foster JA. A comparison of manual kinetic and automated static perimetry in obtaining ptosis fields. Arch Ophthalmol 2000;118:65–9. 6. Lord RK, Shah VA, San Filippo AN, et al. Novel uses of smartphones in ophthalmology. Ophthalmology 2010;117:1274–1274. e3. 7. Alniemi ST, Pang NK, Woog JJ, et al. Comparison of automated and manual perimetry in patients with blepharoptosis. Ophthal Plast Reconstr Surg 2013;29:361–3. 8. Kedar S, Ghate D, Corbett JJ. Visual fields in neuro-ophthalmology. Indian J Ophthalmol 2011;59:103–9. 9. Teghtsoonian M. The judgment of size. Am J Psychol 1965;78:392–402. 10. Krider RE, Raghubir P, Krishna A. Pizzas: pi or square? Psychophysical biases in area comparisons. Marketing Science 2001;20:405–25. 11. Anderson RL, Holds JB. Does anyone know how to differentiate a ‘functional’ defect from a cosmetic one? Arch Ophthalmol 1990;108:1685–6.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
145
The Efficacy of a Novel Mobile Phone Application for Goldmann Ptosis Visual Field Interpretation Robi N. Maamari, B.S.*, Michael V. D’Ambrosio, Ph.D.†, Jeffrey M. Joseph, M.D.*, and Jeremiah P. Tao, M.D., F.A.C.S.* *Gavin Herbert Eye Institute, University of California, Irvine; and †Department of Bioengineering, University of California, Berkeley, California, U.S.A.
Purpose: To evaluate the efficacy of a novel mobile phone application that calculates superior visual field defects on Goldmann visual field charts. Methods: Experimental study in which the mobile phone application and 14 oculoplastic surgeons interpreted the superior visual field defect in 10 Goldmann charts. Percent error of the mobile phone application and the oculoplastic surgeons’ estimates were calculated compared with computer software computation of the actual defects. Precision and time efficiency of the application were evaluated by processing the same Goldmann visual field chart 10 repeated times. Results: The mobile phone application was associated with a mean percent error of 1.98% (95% confidence interval[CI], 0.87%–3.10%) in superior visual field defect calculation. The average mean percent error of the oculoplastic surgeons’ visual estimates was 19.75% (95% CI, 14.39%–25.11%). Oculoplastic surgeons, on average, underestimated the defect in all 10 Goldmann charts. There was high interobserver variance among oculoplastic surgeons. The percent error of the 10 repeated measurements on a single chart was 0.93% (95% CI, 0.40%–1.46%). The average time to process 1 chart was 12.9 seconds (95% CI, 10.9–15.0 seconds). Conclusions: The mobile phone application was highly accurate, precise, and time-efficient in calculating the percent superior visual field defect using Goldmann charts. Oculoplastic surgeon visual interpretations were highly inaccurate, highly variable, and usually underestimated the field vision loss. (Ophthal Plast Reconstr Surg 2014;30:141–145)
V
isual field (VF) testing is commonly included in the preoperative evaluation for functional upper eyelid lifting surgery (e.g., ptosis repair, blepharoplasty, brow lift).1 The specific aim of this test is to quantify the amount of superior visual field (SVF) obstruction by comparing results with the eyelids taped or manually elevated to results with the eyelids in the baseline position. The Center for Medicare Services and other thirdparty payers have defined a 30% loss in the overall SVF as a threshold for medical necessity.2,3 Accepted for publication August 30, 2013. Presented at the American Society of Ophthalmic Plastic and Reconstructive Surgery Fall Scientific Symposium, November 2013, New Orleans, Louisiana. Supported, in part, by an institutional Research to Prevent Blindness grant. The authors have no financial or conflicts of interest to disclose. Address correspondence and reprint requests to Robi N. Maamari, Department of Ophthalmology, Gavin Herbert Eye Institute, University of California, Irvine, CA. E-mail: [email protected] DOI: 10.1097/IOP.0000000000000030
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
While straightforward on static, automated (e.g., Humphrey) VF tests that provide a numeric score of number of stimuli seen by the patient and number of stimuli presented, the percent SVF defect on a kinetic manual (Goldmann) ptosis VF test may be problematic because calculations require one to determine area under curvilinear, noncircular, and often irregular curves. One commonly used method to quantify the overall Goldmann SVF defect is to visually estimate the amount of change in the taped and untaped VF. Meyer et al.4 and Riemann et al.5 described using the height of the SVF at the 90° meridian to determine percent SVF impairment; however, this calculation is predictably inaccurate because the area in question is not rectangular, nor perfectly circular, as the simple mathematical formula assumes. Moreover, many patients have disproportionate superonasal or superotemporal SVF loss. The authors are unaware of any standardized tool, software, or device that calculates the percent VF obstruction on Goldmann VF testing. Mobile smart phone and portable tablet devices are becoming more integrated in medical practice and may be leveraged as a tool for these SVF defect calculations.6 This study introduces a novel mobile phone application that calculates percent SVF defects using a Goldmann VF test and evaluates its efficacy compared with oculoplastic surgeons’ visual estimates and the actual computer-calculated values.
METHODS Questionnaires. Fourteen practicing ophthalmology board certified and fellowship-trained oculoplastic surgeons who routinely use Goldmann VF testing in their practice to evaluate for ptosis repair completed a web-based questionnaire that comprises 10 representative Goldmann ptosis SVF, each with 2 drawn curves: a lower curve representing the patient SVF at baseline (untaped eyelid); and an upper curve representing the patient SVF with the eyelid elevated (usually taped). The baseline SVF area was defined as the area delineated by the untaped VF curve and the horizontal axis (0° and 180° meridian). The unobstructed SVF was defined as the area delineated by the taped VF curve and the horizontal axis. Oculoplastic surgeons were instructed to use their method in practice to visually estimate the percentage of SVF defect, defined as the difference between the unobstructed and baseline SVF areas divided by the unobstructed SVF area. Mobile Phone Application. An iPhone (Apple Inc., Cupertino, CA, U.S.A.) application (app) was created to calculate SVF defect percentages. The app uses an iPhone 4S camera to capture images of the Goldmann SVF chart. Morphological image processing techniques detected the taped and untaped curves from the captured image. Color threshold differences were used to detect and distinguish the 2 curves, so the untaped and taped curves were drawn using 2 different colors (red, blue, or
141
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
R. N. Maamari et al.
FIG. 1. A, Screen shot of the application in preview mode. The user aligns the cross target with the vertical and horizontal axes and the outer circle to equally encompass the circle of eccentricity. Chart 1 is shown. B, Result screen with calculation of SVF defect. Once the chart is imaged, the application performs an automated calculation of the SVF defect and displays an animated representation of the SVF curves. C, Representative Goldmann visual field of Chart 2. D, Representative Goldmann visual field of Chart 3. green). A reference target was displayed on the camera preview screen that the user was instructed to align with the vertical and horizontal axes of the Goldmann charts (Fig. 1A). Tapping on the touchscreen captured images, and an automated algorithm calculated and displayed the SVF defect percentage and a representative image of the chart (Fig. 1B). The app was tested using the same 10 Goldmann SVF charts in the questionnaire given to the oculoplastic surgeons (Fig. 1C,D). To assess the precision and time efficiency of the app, 1 author (R.M.) imaged 1 chart 10 times. The mean SVF defect measurement and the mean time to obtain results were calculated with 95% CIs. The time to obtain a result was defined as the time elapsed between the start of the app and the display of the results screen. Comparison of Superior Visual Field Defect Calculations. Actual SVF defects were calculated using the open-source ImageJ software (National Institute of Health, MD, U.S.A.). After scanning the Goldmann charts to a computer, ImageJ software was used to manually create
142
geometric shapes from the taped and untaped SVF curves to calculate the areas within, from which the percent defect was then determined. The percent error (defined as the absolute value of the difference between the tested method area and the actual area divided by the actual area) was calculated for all 10 charts for the graders’ estimates and the iPhone app. The mean percent errors with 95% CIs were calculated for the graders’ estimates and the iPhone app calculations.
RESULTS Table 1 shows the responses provided by the oculoplastic surgeons, the measurements calculated by the iPhone app, and the calculated percent errors. Questionnaire. The average of the mean percent error of the oculoplastic surgeons’ visual estimates of SVF defects was 19.75% (95% CI [CI], 14.39–25.11%) and the standard deviation (SD) of the graders’ mean
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
(0.93)
(0.31)
(0.66)
(0.65)
(0.30)
(0.38)
(0.53)
(0.97)
Chart 1
Chart 2
Chart 3
Chart 4
Chart 5
Chart 6
Chart 7
Chart 8
Chart 9
1.98 1.56
App 0.19 (0.59) 1.55 (0.92) 0.87 (0.31) 0.48 (0.66) 0.59 (0.65) 3.99 (0.29) 3.81 (0.37) 2.52 (0.54) 4.26 (0.93) 1.57 (0.59)
19.65 19.76
A 32.39 (0.40) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 7.69 (0.60) 66.53 (0.10) 4.70 (0.40) 4.79 (0.50) 7.59 (0.90) 14.36 (0.50) 15.16 15.03
B 32.39 (0.40) 14.06 (0.80) 3.13 (0.30) 6.18 (0.70) 0.01 (0.65) 39.75 (0.18) 21.48 (0.30) 31.45 (0.36) 0.40 (0.97) 2.77 (0.60) 15.57 15.95
C 1.42 (0.60) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 49.79 (0.15) 4.70 (0.40) 4.79 (0.50) 10.67 (0.87) 2.77 (0.60) 19.71 14.12
D 32.39 (0.40) 24.80 (0.70) 32.19 (0.21) 12.02 (0.58) 19.99 (0.52) 43.09 (0.17) 0.54 (0.38) 6.70 (0.49) 2.46 (0.95) 22.93 (0.45) 25.14 18.61
E 32.39 (0.40) 24.80 (0.70) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 66.53 (0.10) 21.48 (0.30) 4.79 (0.50) 2.46 (0.95) 31.49 (0.40) 46.39 17.34
F 57.74 (0.25) 30.17 (0.65) 67.71 (0.10) 46.91 (0.35) 38.48 (0.40) 66.53 (0.10) 60.74 (0.15) 42.88 (0.30) 12.72 (0.85) 40.05 (0.35) 25.43 18.15
G 32.39 (0.40) 19.43 (0.75) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 66.53 (0.10) 4.70 (0.40) 33.29 (0.70) 7.59 (0.90) 22.93 (0.45) 17.48 14.67
H 32.39 (0.40) 14.06 (0.80) 19.28 (0.25) 8.99 (0.60) 15.38 (0.55) 49.79 (0.15) 4.70 (0.40) 4.79 (0.50) 2.46 (0.95) 22.93 (0.45)
Graders’ estimates
9.63 11.09
I 1.42 (0.60) 3.32 (0.90) 19.28 (0.25) 0.12 (0.66) 7.69 (0.60) 33.05 (0.20) 21.48 (0.30) 4.73 (0.55) 2.46 (0.95) 2.77 (0.60) 14.94 12.51
J 32.39 (0.40) 24.80 (0.70) 35.42 (0.20) 16.57 (0.55) 7.69 (0.60) 16.31 (0.25) 0.54 (0.38) 10.51 (0.47) 2.46 (0.95) 2.77 (0.60) 7.73 4.59
K 7.03 (0.55) 3.32 (0.90) 3.13 (0.30) 6.18 (0.70) 7.69 (0.60) 16.31 (0.25) 12.55 (0.43) 10.44 (0.58) 1.43 (0.96) 9.22 (0.53) 18.06 16.62
L 32.39 (0.40) 3.32 (0.90) 35.42 (0.20) 1.40 (0.65) 23.07 (0.50) 49.79 (0.15) 13.62 (0.33) 4.79 (0.50) 2.46 (0.95) 14.36 (0.50)
24.76 15.49
M 23.93 (0.45) 19.43 (0.75) 35.42 (0.20) 1.40 (0.65) 30.76 (0.45) 56.48 (0.13) 21.48 (0.30) 33.35 (0.35) 7.59 (0.90) 17.79 (0.48)
16.83 13.32
N 1.42 (0.60) 14.06 (0.80) 35.42 (0.20) 8.99 (0.60) 23.07 (0.50) 16.31 (0.25) 21.48 (0.30) 4.79 (0.50) 2.68 (1.00) 40.05 (0.35)
(0.49)
(0.93)
(0.49)
(0.34)
(0.16)
(0.53)
(0.60)
(0.22)
(0.78)
Mean estimates (0.45)
Actual, iPhone app, graders’ estimates. Bolded values, percent error values calculated based on actual SVF defect (reported as percentages); Oculoplastic surgeons’ average mean percent error and standard deviation, 19.75% and 9.29%, respectively; Values in parentheses, visual field defects reported as fractions with 1.00 defined as a 100% defect.
Mean percent error Mean SD
Chart 10 (0.58)
Actual SVF defect (0.59)
TABLE 1. Superior visual field defects
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
Mobile Phone App for Goldmann Visual Field Defect
143
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
R. N. Maamari et al.
TABLE 2. Mobile phone app measurements using single chart
Mean SD
Time
SVF defect
Percent error
15.3 13 15.8 9.6 13.6 11.5 9.1 9.5 14.7 17.2
58.71 58.84 59.04 58.43 59.07 58.14 59.02 58.84 57.82 58.2
0.76 0.55 0.2 1.24 0.16 1.73 0.23 0.54 2.26 1.62
12.9 2.9
58.61 0.44
0.93 0.74
SVF, superior visual field; SD, standard deviation. Time is reported in seconds. SVF defect and percent error are reported as percentages.
percent errors was 9.29%. The mean percent error of the oculoplastic surgeon estimates ranged from 7.73% to 46.39%. The SD of the individual grader estimates ranged from 4.59% to 19.76% (Table 1). Mobile Phone Application. The mean percent error of the iPhone application was 1.98% (95% CI, 0.87%–3.10%) with a SD of 1.56% (Table 1). When comparing 10 measurements taken with the iPhone app using a single chart, the average percent error of the 10 measurements was 0.93% (95% CI, 0.40%–1.46%) with a SD of 0.73%. The percent error of the 10 app measurements of the same chart ranged from 0.16% to 2.26%. The average time for analysis of a single Goldmann SVF chart using the app was 12.9 seconds (95% CI, 10.9–15.0 seconds) with a SD of 2.9 seconds (Table 2).
DISCUSSION SVF testing is a common step in the preoperative evaluation for functional upper eyelid surgery. According to an American Society of Ophthalmic Plastic and Reconstructive Surgery national survey on current ptosis management, 87.4% of participants reported using some form of VF preoperatively.1 In addition, many third-party insurers require formal VF quantification of the field defect in their coverage determination for functional eyelid lifting surgery. Riemann et al.5 reported that Goldmann manual kinetic and Humphrey automated static testing are the most common tests. They found both to be effective in documenting ptosisassociated VF loss. However, they concluded that Goldmann testing was more sensitive and 5 times faster than the Humphrey test. Alniemi et al.7 demonstrated that Goldmann testing remained significantly more time-efficient even compared with current Humphrey software. In addition, they reported that 7 of 10 patients preferred Goldmann over Humphrey testing. Though many clinicians prefer Humphrey automated static testing for ease and objectivity in the general ophthalmology setting, Goldmann perimetry remains a workhorse for ptosis and other VF analysis.7,8 A limitation of Goldmann ptosis VF tests is that the interpretation may be more subjective than automated tests. The included cohort of fellowship-trained oculoplastic surgeons had an average of 19.75% error from the actual SVF defect on Goldmann. Oculoplastic surgeons, on average, underestimated the SVF defect, with 8 of the 10 charts showing a statistically significant underestimation (p < 0.05) (Fig. 2). These findings
144
FIG. 2. Graph demonstrating the mean oculoplastic surgeon estimate of percent superior visual field defect and the computer software calculated (actual) value for the 10 Goldmann charts. Error bars represent the 95% confidence interval. SVF, superior visual field.
suggest that current standards of VF interpretation may underdiagnose functional eyelid ptosis or dermatochalasis, at least by present VF defect standards. The reasons for underestimation are not certain, but the authors hypothesize that human visual estimation bias might play a role. First, figures of the same shape but different sizes are perceived to vary less in size than the actual size change.9 The similarly shaped untaped and taped curves may contribute to this underestimation bias. Second, visual estimations comparing areas of 2 shapes are often made using a dominant single dimension measurement to compare the 2 areas.10 In the Goldmann charts, the 90° superior meridian provides a convenient dominant dimension to compare the areas of the taped and uptaped areas. However, employing Chart 2 as an example (Fig. 1C), utilizing the superior meridian as a dominant dimension (10° untaped and 40° taped), leads to an underestimation of the SVF defect (75% calculating area as a rectangle using the superior meridian as the height as compared with 93.75% when calculating the area as a circle using the superior meridian as the radius). Interestingly, the mean oculoplastic surgeon estimate for this chart was 78%, which approaches the 75% estimate using the superior meridian for the rectangular area, and the actual defect was calculated as 93%, which is better approximated by the circular area estimate. Last, oculoplastic surgeons may have a subconscious bias to be conservative and underestimate the SVF defects to avoid third-party payer medical necessity scrutiny. Further research is necessary to clarify the reasons for underestimation by visual interpretation methods. The app may address the variability in visual SVF defect estimates and the lack of a standardized method for Goldmann test analysis. The iPhone app was shown to be statistically superior to the oculoplastic surgeons’ estimates for the calculation of the SVF defect (p < 0.05). This application accurately calculates the overall SVF defect, whereas other methods only provide estimates. For instance, as Anderson et al.11 described, using the height of the curves at the 90° meridian is inaccurate for most Goldmann ptosis field tests but especially when the curve is irregular or unbalanced such as in lateral skin overhang or lateral ptosis with more obstruction temporally.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
Ophthal Plast Reconstr Surg, Vol. 30, No. 2, 2014
With mobile smartphone and portable tablet devices becoming more integrated in medical practice, the percent field defectcalculating app may be easily integrated in practice. However, further refinements may improve the usefulness of the app. Some limitations of the study include the relatively small number of oculoplastic surgeons surveyed. Although the app is highly intuitive and user-friendly, there may be a small learning curve. However, an underestimation bias does not appear to affect this study as the times and the percent errors recorded in Table 2 did not show evidence of improvement in the chronologically listed measurements. The accuracy and precision of the app should, in theory, be equivalent to the computer software values. However, since the app requires the user to align the target to the Goldmann chart, misalignment and varying degrees of camera tilt result in some variance that forestalls perfect precision and accuracy. Nevertheless, Table 3 demonstrates that the 10 measurements of the same chart had only a 0.93% error even with the variations in tilt and misalignment during each individual measurement. Another limitation of the app is the required colorcoded VF curves. Future versions will attempt to eliminate this requirement to increase the ease-of-use of the app. One technological limitation of this study is that the app was only designed to run on the iPhone operating system. Versions for other widely used mobile phone operating systems, such as Android (Google, Inc., Mountain View, CA, U.S.A.), are in the next phases of development. Development of the app for other operating system will be rapidly completed, as much of the app code can be used across various platforms. Further pilot testing may be necessary to ensure smooth and robust functionality. At the completion of pilot testing, the app will be released as a free iPhone app. In summary, this study demonstrates that current methods for SVF defect estimation on ptosis Goldmann VFs may be highly variable and underestimate the actual amount of ptosis. This study also introduces a mobile phone application that can rapidly calculate both precise and accurate SVF defect measurements.
Mobile Phone App for Goldmann Visual Field Defect
REFERENCES 1. Aakalu VK, Setabutr P. Current ptosis management: a national survey of ASOPRS members. Ophthal Plast Reconstr Surg 2011;27:270–6. 2. Medicare Local Coverage Determination for Blepharoplasty, Blepharoptosis, and Brow Lift (Effective 8/18/2008–5/1/2013). Available at: http://www.cms.gov/medicare-coverage-database/details/lcd-details.aspx?LCDId=27474&ContrId=161&ver=59&Con trVer=2&Date=08%2f26%2f2013&DocID=L27474&bc=AAAAA AgAIAAAAA%3d%3d&. Accessed June 27, 2013. 3. United Heath Care Services, Inc. Blepharoplasty, Blepharoptosis, and Brow Ptosis Repair Coverage Determination Guidelines (Effective 4/1/2013). Available at: https://www.unitedhealthcareonline.com/ ccmcontent/ProviderII/UHC/en-US/Assets/ProviderStaticFiles/ ProviderStaticFilesPdf/Tools%20and%20Resources/Policies%20 and%20Protocols/Medical%20Policies/Medical%20Policies/ Blepharoplasty_CD.pdf. Acessed June 27, 2013. 4. Meyer DR, Linberg JV, Powell SR, et al. Quantitating the superior visual field loss associated with ptosis. Arch Ophthalmol 1989;107:840–3. 5. Riemann CD, Hanson S, Foster JA. A comparison of manual kinetic and automated static perimetry in obtaining ptosis fields. Arch Ophthalmol 2000;118:65–9. 6. Lord RK, Shah VA, San Filippo AN, et al. Novel uses of smartphones in ophthalmology. Ophthalmology 2010;117:1274–1274. e3. 7. Alniemi ST, Pang NK, Woog JJ, et al. Comparison of automated and manual perimetry in patients with blepharoptosis. Ophthal Plast Reconstr Surg 2013;29:361–3. 8. Kedar S, Ghate D, Corbett JJ. Visual fields in neuro-ophthalmology. Indian J Ophthalmol 2011;59:103–9. 9. Teghtsoonian M. The judgment of size. Am J Psychol 1965;78:392–402. 10. Krider RE, Raghubir P, Krishna A. Pizzas: pi or square? Psychophysical biases in area comparisons. Marketing Science 2001;20:405–25. 11. Anderson RL, Holds JB. Does anyone know how to differentiate a ‘functional’ defect from a cosmetic one? Arch Ophthalmol 1990;108:1685–6.
© 2014 The American Society of Ophthalmic Plastic and Reconstructive Surgery, Inc.
145
