Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size

Tapan Patel, N. Venkatesh Prajna, Sina Farsiu, Nita G. Valikodath, Leslie M. Niziol, Lakshey Dudeja, Kyeong Hwan Kim, Maria A. Woodward

Research output: Contribution to journalArticle

Abstract

Purpose: To assess variability in corneal ulcer measurements between ophthalmologists and reduce clinician-dependent variability using semiautomated segmentation of the ulcer from photographs. Methods: Three ophthalmologists measured 50 patients' eyes for epithelial defects (EDs) and the stromal infiltrate (SI) size using slit-lamp (SL) calipers. SL photographs were obtained. An algorithm was developed for semiautomatic segmenting of the ED and SI in the photographs. Semiautomatic segmentation was repeated 3 times by different users (2 ophthalmologists and 1 trainee). Clinically significant variability was assessed with intraclass correlation coefficients (ICCs) and the percentage of pairwise measurements differing by ≥0.5 mm. Semiautomatic segmentation measurements were compared with manual delineation of the image by a corneal specialist (gold standard) using Dice similarity coefficients. Results: Ophthalmologists' reliability in measurements by SL calipers had an ICC from 0.84 to 0.88 between examiners. Measurements by semiautomatic segmentation had an ICC from 0.96 to 0.98. SL measures of ulcers by clinical versus semiautomatic segmentation measures differed by ≥0.5 mm in 24% to 38% versus 8% to 28% (ED height); 30% to 52% versus 12% to 34% (ED width); 26% to 38% versus 10% to 32% (SI height); and 38% to 58% versus 14% to 34% (SI width), respectively. Average Dice similarity coefficients between manual and repeated semiautomatic segmentation ranged from 0.83 to 0.86 for the ED and 0.78 to 0.83 for the SI. Conclusions: Variability exists when measuring corneal ulcers, even among ophthalmologists. Photography and computerized methods for quantifying the ulcer size could reduce variability while remaining accurate and impact quantitative measurement endpoints.

Original languageEnglish (US)
Pages (from-to)331-339
Number of pages9
JournalCornea
Volume37
Issue number3
DOIs
StatePublished - Mar 1 2018
Externally publishedYes

Fingerprint

Corneal Ulcer
Ulcer
Photography
Reproducibility of Results
Ophthalmologists
Slit Lamp

Keywords

  • corneal ulcer
  • interexaminer variability
  • random forest segmentation
  • semiautomated measurement

ASJC Scopus subject areas

  • Ophthalmology

Cite this

Patel, T., Prajna, N. V., Farsiu, S., Valikodath, N. G., Niziol, L. M., Dudeja, L., ... Woodward, M. A. (2018). Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size. Cornea, 37(3), 331-339. https://doi.org/10.1097/ICO.0000000000001488

Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size. / Patel, Tapan; Prajna, N. Venkatesh; Farsiu, Sina; Valikodath, Nita G.; Niziol, Leslie M.; Dudeja, Lakshey; Kim, Kyeong Hwan; Woodward, Maria A.

In: Cornea, Vol. 37, No. 3, 01.03.2018, p. 331-339.

Research output: Contribution to journalArticle

Patel, T, Prajna, NV, Farsiu, S, Valikodath, NG, Niziol, LM, Dudeja, L, Kim, KH & Woodward, MA 2018, 'Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size', Cornea, vol. 37, no. 3, pp. 331-339. https://doi.org/10.1097/ICO.0000000000001488
Patel, Tapan ; Prajna, N. Venkatesh ; Farsiu, Sina ; Valikodath, Nita G. ; Niziol, Leslie M. ; Dudeja, Lakshey ; Kim, Kyeong Hwan ; Woodward, Maria A. / Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size. In: Cornea. 2018 ; Vol. 37, No. 3. pp. 331-339.
@article{cb9f2d68bddd4bdc9958afd53c2a8567,
title = "Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size",
abstract = "Purpose: To assess variability in corneal ulcer measurements between ophthalmologists and reduce clinician-dependent variability using semiautomated segmentation of the ulcer from photographs. Methods: Three ophthalmologists measured 50 patients' eyes for epithelial defects (EDs) and the stromal infiltrate (SI) size using slit-lamp (SL) calipers. SL photographs were obtained. An algorithm was developed for semiautomatic segmenting of the ED and SI in the photographs. Semiautomatic segmentation was repeated 3 times by different users (2 ophthalmologists and 1 trainee). Clinically significant variability was assessed with intraclass correlation coefficients (ICCs) and the percentage of pairwise measurements differing by ≥0.5 mm. Semiautomatic segmentation measurements were compared with manual delineation of the image by a corneal specialist (gold standard) using Dice similarity coefficients. Results: Ophthalmologists' reliability in measurements by SL calipers had an ICC from 0.84 to 0.88 between examiners. Measurements by semiautomatic segmentation had an ICC from 0.96 to 0.98. SL measures of ulcers by clinical versus semiautomatic segmentation measures differed by ≥0.5 mm in 24{\%} to 38{\%} versus 8{\%} to 28{\%} (ED height); 30{\%} to 52{\%} versus 12{\%} to 34{\%} (ED width); 26{\%} to 38{\%} versus 10{\%} to 32{\%} (SI height); and 38{\%} to 58{\%} versus 14{\%} to 34{\%} (SI width), respectively. Average Dice similarity coefficients between manual and repeated semiautomatic segmentation ranged from 0.83 to 0.86 for the ED and 0.78 to 0.83 for the SI. Conclusions: Variability exists when measuring corneal ulcers, even among ophthalmologists. Photography and computerized methods for quantifying the ulcer size could reduce variability while remaining accurate and impact quantitative measurement endpoints.",
keywords = "corneal ulcer, interexaminer variability, random forest segmentation, semiautomated measurement",
author = "Tapan Patel and Prajna, {N. Venkatesh} and Sina Farsiu and Valikodath, {Nita G.} and Niziol, {Leslie M.} and Lakshey Dudeja and Kim, {Kyeong Hwan} and Woodward, {Maria A.}",
year = "2018",
month = "3",
day = "1",
doi = "10.1097/ICO.0000000000001488",
language = "English (US)",
volume = "37",
pages = "331--339",
journal = "Cornea",
issn = "0277-3740",
publisher = "Lippincott Williams and Wilkins",
number = "3",

}

TY - JOUR

T1 - Novel Image-Based Analysis for Reduction of Clinician-Dependent Variability in Measurement of the Corneal Ulcer Size

AU - Patel, Tapan

AU - Prajna, N. Venkatesh

AU - Farsiu, Sina

AU - Valikodath, Nita G.

AU - Niziol, Leslie M.

AU - Dudeja, Lakshey

AU - Kim, Kyeong Hwan

AU - Woodward, Maria A.

PY - 2018/3/1

Y1 - 2018/3/1

N2 - Purpose: To assess variability in corneal ulcer measurements between ophthalmologists and reduce clinician-dependent variability using semiautomated segmentation of the ulcer from photographs. Methods: Three ophthalmologists measured 50 patients' eyes for epithelial defects (EDs) and the stromal infiltrate (SI) size using slit-lamp (SL) calipers. SL photographs were obtained. An algorithm was developed for semiautomatic segmenting of the ED and SI in the photographs. Semiautomatic segmentation was repeated 3 times by different users (2 ophthalmologists and 1 trainee). Clinically significant variability was assessed with intraclass correlation coefficients (ICCs) and the percentage of pairwise measurements differing by ≥0.5 mm. Semiautomatic segmentation measurements were compared with manual delineation of the image by a corneal specialist (gold standard) using Dice similarity coefficients. Results: Ophthalmologists' reliability in measurements by SL calipers had an ICC from 0.84 to 0.88 between examiners. Measurements by semiautomatic segmentation had an ICC from 0.96 to 0.98. SL measures of ulcers by clinical versus semiautomatic segmentation measures differed by ≥0.5 mm in 24% to 38% versus 8% to 28% (ED height); 30% to 52% versus 12% to 34% (ED width); 26% to 38% versus 10% to 32% (SI height); and 38% to 58% versus 14% to 34% (SI width), respectively. Average Dice similarity coefficients between manual and repeated semiautomatic segmentation ranged from 0.83 to 0.86 for the ED and 0.78 to 0.83 for the SI. Conclusions: Variability exists when measuring corneal ulcers, even among ophthalmologists. Photography and computerized methods for quantifying the ulcer size could reduce variability while remaining accurate and impact quantitative measurement endpoints.

AB - Purpose: To assess variability in corneal ulcer measurements between ophthalmologists and reduce clinician-dependent variability using semiautomated segmentation of the ulcer from photographs. Methods: Three ophthalmologists measured 50 patients' eyes for epithelial defects (EDs) and the stromal infiltrate (SI) size using slit-lamp (SL) calipers. SL photographs were obtained. An algorithm was developed for semiautomatic segmenting of the ED and SI in the photographs. Semiautomatic segmentation was repeated 3 times by different users (2 ophthalmologists and 1 trainee). Clinically significant variability was assessed with intraclass correlation coefficients (ICCs) and the percentage of pairwise measurements differing by ≥0.5 mm. Semiautomatic segmentation measurements were compared with manual delineation of the image by a corneal specialist (gold standard) using Dice similarity coefficients. Results: Ophthalmologists' reliability in measurements by SL calipers had an ICC from 0.84 to 0.88 between examiners. Measurements by semiautomatic segmentation had an ICC from 0.96 to 0.98. SL measures of ulcers by clinical versus semiautomatic segmentation measures differed by ≥0.5 mm in 24% to 38% versus 8% to 28% (ED height); 30% to 52% versus 12% to 34% (ED width); 26% to 38% versus 10% to 32% (SI height); and 38% to 58% versus 14% to 34% (SI width), respectively. Average Dice similarity coefficients between manual and repeated semiautomatic segmentation ranged from 0.83 to 0.86 for the ED and 0.78 to 0.83 for the SI. Conclusions: Variability exists when measuring corneal ulcers, even among ophthalmologists. Photography and computerized methods for quantifying the ulcer size could reduce variability while remaining accurate and impact quantitative measurement endpoints.

KW - corneal ulcer

KW - interexaminer variability

KW - random forest segmentation

KW - semiautomated measurement

UR - http://www.scopus.com/inward/record.url?scp=85045536551&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045536551&partnerID=8YFLogxK

U2 - 10.1097/ICO.0000000000001488

DO - 10.1097/ICO.0000000000001488

M3 - Article

VL - 37

SP - 331

EP - 339

JO - Cornea

JF - Cornea

SN - 0277-3740

IS - 3

ER -