Purchase this article with an account.
Makoto Saika, Naoyuki Maeda, Tomoya Nakagawa, Yoko Hirohara, Takashi Fujikado, Toshifumi Mihashi; The Keratoconus screening using logistic regression analysis of Corneal topography. Journal of Vision 2008;8(17):79. doi: https://doi.org/10.1167/8.17.79.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Keratoconus is a noninflammatory disease by the morphological changes of corneal shape. With the progress of the corneal shape measurement and the development of the computer, many papers are reported form 1990's. Almost these papers are used Decision Tree Regression(DT) and Neural Network(NN) as an automated pattern classification. So we compared the other statics regression, Linear Discriminant Analysis(LDA) and k-Nearest Neighbor(kNN), with DT and NN at ARVO2008. In this study, we examined a Logistic Regression (LR) using Akaike Information Criterion (AIC) to evaluate suitability of the regression function and to compare it to the other classifications. For the subjects of 51 keratoconic eyes (KC), 46 keratoconus suspect eyes (KCS) and 65 normal eyes (N), Placido-based videokeratography was performed with KR-9000PW (Topcon) for the anterior corneal surface, and the corneal wavefront aberrations were analyzed with the Zernike polynomials and some indexes of corneal topography. One half of the subjects were randomly chosen as the training set and the other half were used as the test set. Using the training set, the LR was applied to achieve automated classification and the sensitivity and the specificity were 96% and 94%. Precision of the LR was better than regression analyses in the previous studies. Evaluating the LR by using AIC showed that Keratoconus is characterized by six Zernike modes, , and corneal radius of curvature (simulated cornea).
This PDF is available to Subscribers Only