advertisement
Editors Selection IGR 8-4
Imaging: Automated analysis
Comment by Linda Zangwill on:
15098 Automated analysis of heidelberg retina tomograph optic disc images by glaucoma probability score, Coops A; Henson DB; Kwartz AJ et al., Investigative Ophthalmology and Visual Science, 2006; 47: 5348-5355
Improvements in optical imaging instruments have enhanced their ability to provide objective information that can assist the clinician in diagnosing and monitoring glaucomatous structural changes. Coops et al. (1110) evaluated the diagnostic accuracy of the Glaucoma Probability Score (GPS), a new analysis available with Heidelberg Retina Tomograph (HRT) 3.0 software that calculates the likelihood of glaucoma damage (as a probability from 0% to 100%), without the need to manually draw a contour line to outline the optic disc. The GPS uses three features of the optic disc and two measures of the retinal nerve fiber layer (RNFL) in a sophisticated machine learning classifier that compares the model to those obtained in healthy and glaucomatous eyes. The GPS therefore provides a fully automated analysis of the optic disc as a global score and by region in a manner similar to that of the Moorfields Regression Analysis (MRA), an analysis routinely utilized by clinicians that requires the outlining of the disc margin. Coop et al. provide a comprehensive and thorough comparison of the GPS and MRA. The authors conclude that the diagnostic accuracy of the GPS is similar to the MRA (areas under the receiver operating characteristic (ROC) curve of .78 and .77, respectively), and that there is good agreement between methods, (71% of glaucoma eyes and 68% of healthy eyes). In addition, they present a proportional odds logistic regression to quantitatively simultaneous control for several covariates.
The diagnostic accuracy of the Glaucoma Probability Score is similar to the Moorfield Regression Analysis and there is good agreement between methodsOf particular interest is optic disc size, as it has been shown to influence the diagnostic accuracy of HRT parameters including MRA. However, as the GPS calculation is based on the overall shape of the optic nerve head and does not rely on the outlining of the disc margin for its calculation, it may be less influenced by optic disc size than the MRA. The authors report that disc size influences the diagnostic accuracy of the GPS, in a manner similar to MRA; an estimated 21% and 15% increase, respectively, in the odds of a positive classification for each .1 mm2 increase in disc area. The authors conclude that the sensitivity increases and specificity decreases with increasing disc size in both GPS and MRA analysis and appropriately highlight the possibility of a high false positive rate in eyes with large discs. Previous studies have found that both between disease severity and disc size influence diagnostic accuracy of HRT parameters. It is therefore surprising that in this study, disease severity as measured by visual field indices MD and PSD did not influence the classification outcome of either MRA or GPS. The authors do not offer an explanation for this result.
As the authors mention in the discussion, 'the Standards for the Reporting of Diagnostic Accuracy (STARD) have called for the complete reporting of relevant covariates (e.g., measures of disease severity) that may influence diagnostic performance, reporting of such details has been incomplete in many previous studies on the HRT.'