Skip to main content

Table 3 Performance of the discriminating rules generated using logistic regression and neural network classifiers for differentiating sub KC and KC corneas from normal corneas

From: Machine learning helps improve diagnostic ability of subclinical keratoconus using Scheimpflug and OCT imaging modalities

  Normal vs. Sub KC Normal vs. KC
  Sensitivity 1-Specificity AUC Sensitivity 1-Specificity AUC
Logistic Regression       
Pentacam HR system 83.8% 88.7% 0.74 100% 100% 1.00
UHR-OCT 95.3% 94.5% 0.90 98.0% 100% 0.98
Pentacam HR system
& UHR-OCT
95.1% 94.8% 0.90 100% 99.4% 0.99
Neural Network       
Pentacam HR system 82.1% 82.6% 0.68 100% 100% 1.00
UHR-OCT 94.8% 93.4% 0.88 99.5% 99% 0.98
Pentacam HR system
& UHR-OCT
98.5% 94.7% 0.93 100% 100% 1.00
  1. UHR-OCT= ultra-high resolution optical coherence tomography; Sub KC= subclinical keratoconus group; KC= keratoconus group