ABSTRACT
Purpose: We compared the characteristics of subtle morphological changes in subclinical keratoconus (KC) and normal corneas using Scheimpflug tomography (Pentacam®) and assessed the efficacy of these parameters for distinguishing KC or subclinical KC from normal eyes. Patients and Methods: In this multicenter comparative study at Dhahran Eye Specialist Hospital and Al Kahhal Medical Complex in the Eastern Province of Saudi Arabia, we analyzed the Scheimpflug tomography charts of patients with topographically normal eyes and those with unilateral KC. Patients were divided into the normal (NL: patients considered for refractive surgery and with normal topographic/tomographic features, 129 eyes), KC (30 patients with manifest KC in one eye based on biomicroscopy and topographical findings), and forme fruste KC (FFKC: fellow eyes of patients in the KC group that met the NL group criteria) groups. Corneal morphological parameters were analyzed using the area under the receiver operating characteristic (ROC) curves (AUCs). Results: For distinguishing NL and KC groups, all measured corneal morphological parameters, except for flat keratometry, maximum Ambrósio relational thickness index, and minimum sagittal curvature, had AUCs >0.75. The surface variance index yielded the largest AUC (0.999). For distinguishing NL and FFKC groups, all corneal morphological parameters had AUCs <0.8. Total higher-order aberrations (RMS HOA) yielded the highest AUC, followed by Belin/Ambrosio Enhanced Ectasia total deviation (BAD-D), back elevation at the thinnest location, average pachymetric progression index (PPIave), and deviation of Ambrosio relational thickness (Da) (AUC 0.74-0.78). Conclusion: The diagnostic performance of all tested topographic and tomographic parameters measured using Scheimpflug tomography for discriminating subclinical KC was fair at best, with the top parameters being RMS HOA, BAD-D, back elevation at the thinnest location, PPIave, and Da. Distinguishing between subclinical KC and healthy eyes remains challenging. Multimodal imaging techniques may be required for optimal early detection of subtle morphological changes.
ABSTRACT
Purpose: To compare supervised transfer learning to semisupervised learning for their ability to learn in-depth knowledge with limited data in the optical coherence tomography (OCT) domain. Methods: Transfer learning with EfficientNet-B4 and semisupervised learning with SimCLR are used in this work. The largest public OCT dataset, consisting of 108,312 images and four categories (choroidal neovascularization, diabetic macular edema, drusen, and normal) is used. In addition, two smaller datasets are constructed, containing 31,200 images for the limited version and 4000 for the mini version of the dataset. To illustrate the effectiveness of the developed models, local interpretable model-agnostic explanations and class activation maps are used as explainability techniques. Results: The proposed transfer learning approach using the EfficientNet-B4 model trained on the limited dataset achieves an accuracy of 0.976 (95% confidence interval [CI], 0.963, 0.983), sensitivity of 0.973 and specificity of 0.991. The semisupervised based solution with SimCLR using 10% labeled data and the limited dataset performs with an accuracy of 0.946 (95% CI, 0.932, 0.960), sensitivity of 0.941, and specificity of 0.983. Conclusions: Semisupervised learning has a huge potential for datasets that contain both labeled and unlabeled inputs, generally, with a significantly smaller number of labeled samples. The semisupervised based solution provided with merely 10% labeled data achieves very similar performance to the supervised transfer learning that uses 100% labeled samples. Translational Relevance: Semisupervised learning enables building performant models while requiring less expertise effort and time by using to good advantage the abundant amount of available unlabeled data along with the labeled samples.
