RESUMO
The morphologic appearance of the optic disc is of interest in glaucoma. In contrast to descriptive classification systems that are currently used, a quantitative approach to the analysis of optic disc morphology is required. Our goal was to determine the optimal method for quantifying optic cup shape by comparing traditional (ovality, form-factor and neuroretinal rim (NRR) width ratio) and geometric morphometric approaches. Left optic disc stereophotographs of 160 (80 normal and 80 glaucomatous (stratified by severity)) subjects were examined. The optic cup margins were stereoscopically delineated with a custom tracing system and saved as a series of discrete points. The geometric morphometric methods of elliptic Fourier analysis (EFA) and sliding semi-landmark analysis (SSLA) were used to eliminate variation unrelated to shape (e.g. size) and yield a series of shape variables. Differences in optic cup shape between normal and glaucoma groups were investigated. Discriminant functions were computed and the sensitivity and specificity of each technique determined. Receiver operator characteristic (ROC) curves were calculated for all methods and evaluated in their potential to discriminate between normal and glaucomatous eyes based on the shape variables. All geometric morphometric methods revealed differences between normal and glaucomatous eyes in optic cup shape, in addition to the traditional parameters of ovality, form-factor and NRR width ratio (p<0.0005). SSLA (minimum bending energy criterion--18 points) had the best sensitivity (83%) and area under the curve (AUC) (0.91). EFA (72 points) performed similarly well (74%, 0.89) as did the set of traditional shape-based variables (76%, 0.86). This study demonstrated that a geometric morphometric approach for discriminating between normal and glaucomatous eyes in optic cup shape is superior to that provided by traditional single parameter shape measures. Such analytical techniques could be incorporated into future automated optic disc screening modalities.