Atypical retardation pattern: Can performance of classification be improved?

Pinakin Gunvant, Yufeng Zheng, Marta Tóth, G. Holló

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

PURPOSE.: (1) To evaluate and compare the classification performance of Wavelet-Fourier analysis (WFA), Fast-Fourier analysis (FFA), and the standard GDx-variable corneal compensator (VCC) output in identifying glaucomatous eyes from a mixed group of healthy and glaucomatous eyes with atypical retardation pattern (ARP). (2) To investigate if classification performance improves when only the superior and inferior quadrants are used for WFA and FFA. (3) To evaluate the classification performance as a function of severity of ARP. METHODS.: Retinal nerve fiber layer (RNFL) estimates were obtained from 445 eyes of 240 individuals. On the basis of typical scan score (TSS), 348 eyes had typical retardation pattern (TRP) and 97 had ARP (78% TRP and 22% ARP). The classification performance of WFA and FFA classifiers was tested using three different ways: (1) Classifiers were trained on the TRP data, and tested on ARP data. (2) Classifiers were trained on TRP and 90% ARP data using 10-fold cross validation technique and tested on ARP data (10%). (3) Classifiers were trained and tested using the ARP data using 10-fold cross validation technique. Sensitivity, specificity, and Receiver Operating Characteristic Curve (ROC) areas were calculated. The classification performance was also assessed for the standard parameters of GDx-VCC. RESULTS.: Of the standard GDx-VCC parameters, the nerve fiber indicator (NFI) had the highest ROC area (0.80). Of the shape-based analyses, WFA and FFA of the complete temporal, superior, nasal, inferior, and temporal curve had the highest ROC area (0.85 and 0.82, respectively). The difference in the ROC areas did not reach the statistically significant level (p = 0.07). On eyes with severe ARP (TSS <60) all metrics performed similarly, but in case of moderate ARP (TSS 60 to 79), the ROC area of WFA and FFA were both greater than that of NFI (the difference was 9% and 7%, respectively). CONCLUSION.: Although the WFA and FFA classification performance was greater than NFI as assessed by ROC area the difference was not statistically significant.

Original languageEnglish
JournalOptometry and Vision Science
Volume85
Issue number6
DOIs
Publication statusPublished - Jun 2008

Fingerprint

Fourier Analysis
Wavelet Analysis
ROC Curve
Nerve Fibers
Nose
Sensitivity and Specificity

Keywords

  • Atypical retardation pattern
  • Fourier analysis
  • Polarimetry
  • Retinal nerve fiber layer
  • Wavelet-Fourier analysis

ASJC Scopus subject areas

  • Ophthalmology
  • Optometry

Cite this

Atypical retardation pattern : Can performance of classification be improved? / Gunvant, Pinakin; Zheng, Yufeng; Tóth, Marta; Holló, G.

In: Optometry and Vision Science, Vol. 85, No. 6, 06.2008.

Research output: Contribution to journalArticle

Gunvant, Pinakin ; Zheng, Yufeng ; Tóth, Marta ; Holló, G. / Atypical retardation pattern : Can performance of classification be improved?. In: Optometry and Vision Science. 2008 ; Vol. 85, No. 6.
@article{5a0c561a9fd147698249262d4a3c6153,
title = "Atypical retardation pattern: Can performance of classification be improved?",
abstract = "PURPOSE.: (1) To evaluate and compare the classification performance of Wavelet-Fourier analysis (WFA), Fast-Fourier analysis (FFA), and the standard GDx-variable corneal compensator (VCC) output in identifying glaucomatous eyes from a mixed group of healthy and glaucomatous eyes with atypical retardation pattern (ARP). (2) To investigate if classification performance improves when only the superior and inferior quadrants are used for WFA and FFA. (3) To evaluate the classification performance as a function of severity of ARP. METHODS.: Retinal nerve fiber layer (RNFL) estimates were obtained from 445 eyes of 240 individuals. On the basis of typical scan score (TSS), 348 eyes had typical retardation pattern (TRP) and 97 had ARP (78{\%} TRP and 22{\%} ARP). The classification performance of WFA and FFA classifiers was tested using three different ways: (1) Classifiers were trained on the TRP data, and tested on ARP data. (2) Classifiers were trained on TRP and 90{\%} ARP data using 10-fold cross validation technique and tested on ARP data (10{\%}). (3) Classifiers were trained and tested using the ARP data using 10-fold cross validation technique. Sensitivity, specificity, and Receiver Operating Characteristic Curve (ROC) areas were calculated. The classification performance was also assessed for the standard parameters of GDx-VCC. RESULTS.: Of the standard GDx-VCC parameters, the nerve fiber indicator (NFI) had the highest ROC area (0.80). Of the shape-based analyses, WFA and FFA of the complete temporal, superior, nasal, inferior, and temporal curve had the highest ROC area (0.85 and 0.82, respectively). The difference in the ROC areas did not reach the statistically significant level (p = 0.07). On eyes with severe ARP (TSS <60) all metrics performed similarly, but in case of moderate ARP (TSS 60 to 79), the ROC area of WFA and FFA were both greater than that of NFI (the difference was 9{\%} and 7{\%}, respectively). CONCLUSION.: Although the WFA and FFA classification performance was greater than NFI as assessed by ROC area the difference was not statistically significant.",
keywords = "Atypical retardation pattern, Fourier analysis, Polarimetry, Retinal nerve fiber layer, Wavelet-Fourier analysis",
author = "Pinakin Gunvant and Yufeng Zheng and Marta T{\'o}th and G. Holl{\'o}",
year = "2008",
month = "6",
doi = "10.1097/OPX.0b013e3181783aa2",
language = "English",
volume = "85",
journal = "Optometry and Vision Science",
issn = "1040-5488",
publisher = "Lippincott Williams and Wilkins",
number = "6",

}

TY - JOUR

T1 - Atypical retardation pattern

T2 - Can performance of classification be improved?

AU - Gunvant, Pinakin

AU - Zheng, Yufeng

AU - Tóth, Marta

AU - Holló, G.

PY - 2008/6

Y1 - 2008/6

N2 - PURPOSE.: (1) To evaluate and compare the classification performance of Wavelet-Fourier analysis (WFA), Fast-Fourier analysis (FFA), and the standard GDx-variable corneal compensator (VCC) output in identifying glaucomatous eyes from a mixed group of healthy and glaucomatous eyes with atypical retardation pattern (ARP). (2) To investigate if classification performance improves when only the superior and inferior quadrants are used for WFA and FFA. (3) To evaluate the classification performance as a function of severity of ARP. METHODS.: Retinal nerve fiber layer (RNFL) estimates were obtained from 445 eyes of 240 individuals. On the basis of typical scan score (TSS), 348 eyes had typical retardation pattern (TRP) and 97 had ARP (78% TRP and 22% ARP). The classification performance of WFA and FFA classifiers was tested using three different ways: (1) Classifiers were trained on the TRP data, and tested on ARP data. (2) Classifiers were trained on TRP and 90% ARP data using 10-fold cross validation technique and tested on ARP data (10%). (3) Classifiers were trained and tested using the ARP data using 10-fold cross validation technique. Sensitivity, specificity, and Receiver Operating Characteristic Curve (ROC) areas were calculated. The classification performance was also assessed for the standard parameters of GDx-VCC. RESULTS.: Of the standard GDx-VCC parameters, the nerve fiber indicator (NFI) had the highest ROC area (0.80). Of the shape-based analyses, WFA and FFA of the complete temporal, superior, nasal, inferior, and temporal curve had the highest ROC area (0.85 and 0.82, respectively). The difference in the ROC areas did not reach the statistically significant level (p = 0.07). On eyes with severe ARP (TSS <60) all metrics performed similarly, but in case of moderate ARP (TSS 60 to 79), the ROC area of WFA and FFA were both greater than that of NFI (the difference was 9% and 7%, respectively). CONCLUSION.: Although the WFA and FFA classification performance was greater than NFI as assessed by ROC area the difference was not statistically significant.

AB - PURPOSE.: (1) To evaluate and compare the classification performance of Wavelet-Fourier analysis (WFA), Fast-Fourier analysis (FFA), and the standard GDx-variable corneal compensator (VCC) output in identifying glaucomatous eyes from a mixed group of healthy and glaucomatous eyes with atypical retardation pattern (ARP). (2) To investigate if classification performance improves when only the superior and inferior quadrants are used for WFA and FFA. (3) To evaluate the classification performance as a function of severity of ARP. METHODS.: Retinal nerve fiber layer (RNFL) estimates were obtained from 445 eyes of 240 individuals. On the basis of typical scan score (TSS), 348 eyes had typical retardation pattern (TRP) and 97 had ARP (78% TRP and 22% ARP). The classification performance of WFA and FFA classifiers was tested using three different ways: (1) Classifiers were trained on the TRP data, and tested on ARP data. (2) Classifiers were trained on TRP and 90% ARP data using 10-fold cross validation technique and tested on ARP data (10%). (3) Classifiers were trained and tested using the ARP data using 10-fold cross validation technique. Sensitivity, specificity, and Receiver Operating Characteristic Curve (ROC) areas were calculated. The classification performance was also assessed for the standard parameters of GDx-VCC. RESULTS.: Of the standard GDx-VCC parameters, the nerve fiber indicator (NFI) had the highest ROC area (0.80). Of the shape-based analyses, WFA and FFA of the complete temporal, superior, nasal, inferior, and temporal curve had the highest ROC area (0.85 and 0.82, respectively). The difference in the ROC areas did not reach the statistically significant level (p = 0.07). On eyes with severe ARP (TSS <60) all metrics performed similarly, but in case of moderate ARP (TSS 60 to 79), the ROC area of WFA and FFA were both greater than that of NFI (the difference was 9% and 7%, respectively). CONCLUSION.: Although the WFA and FFA classification performance was greater than NFI as assessed by ROC area the difference was not statistically significant.

KW - Atypical retardation pattern

KW - Fourier analysis

KW - Polarimetry

KW - Retinal nerve fiber layer

KW - Wavelet-Fourier analysis

UR - http://www.scopus.com/inward/record.url?scp=48549106163&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=48549106163&partnerID=8YFLogxK

U2 - 10.1097/OPX.0b013e3181783aa2

DO - 10.1097/OPX.0b013e3181783aa2

M3 - Article

C2 - 18521014

AN - SCOPUS:48549106163

VL - 85

JO - Optometry and Vision Science

JF - Optometry and Vision Science

SN - 1040-5488

IS - 6

ER -