Purchase this article with an account.
P. Updike, A.C. Walsh, W.Z. Wu, P.W. Romano, S.R. Sadda; Comparison of Computer–Based Digital Grading to Manual Assessment of Fundus Images . Invest. Ophthalmol. Vis. Sci. 2005;46(13):234.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Software has been developed for computer–assisted grading of retinal images, which allows precise measurements of lesion sizes, rather than the categorical assessments made by traditional reading center methods. The purpose of this study is to assess the magnitude of the error introduced by categorical methods.
Ten patients with subfoveal neovascular age–related macular degeneration (NVAMD) underwent treatment with a novel therapy (not the subject of this poster) and were reevaluated three months later. Digital color fundus photographs and flouroscein angiograms (Topcon 50IX, 6Mpixel) were performed according to protocols described for previous AMD clinical trials. Digital images were evaluated by trained human graders at the Doheny Image Reading Center (DIRC) and the CNV lesions sizes were assessed according to Macular Photocoagulation Study (MPS) disk area (DA) circles [0, <=1, <=2, <=3, <=4, <=5, <=6, <=9, <=12, 16]. Assessments were also performed using DIRC software designed to allow graders to precisely outline and identify all lesion components. Areas of the outlined lesion were calculated by the computer and expressed in precise MPS disk areas.
The baseline CNV lesion sizes, follow–up lesion size, and change in lesion size (baseline to month 3) are shown in Table 1 for both grading methods. The percent error between the traditional categorical versus computer–assisted measurements is also shown. An error of at least 20% was noted in 4/10 cases including one case with a difference of 2.5 MPS DA, in which the categorical system was completely insensitive for detecting a significant change in the lesion size.
Significant errors can occur when using categorical MPS DA circles for assessing lesion size. Such errors may not be significant for large–scale clinical trials but their importance may be magnified in smaller studies designed to evaluate safety and efficacy at an early stage. Computer–assisted tools for precise measurement of lesions may help address this problem.
This PDF is available to Subscribers Only