From:  Multimodal feature extraction and fusion for determining RGP lens specification base-curve through Pentacam images

 Summary of previous research with the best method and result.

ResearchSubjectMethodResult
[2]Determining the base curve by manual and device-based methodsCalculens algorithm Success rate 50.6%
[6]Determining the base curve based on CAE featuresBasic CAE and pre-trained CAE on the features of Pentacam imagesMSE = 0.056
[4]Image processing/machine learning-based RGP lens base-curve determination on Pentacam four refractive mapsClassical image processing method with combinatory feature fusionMSE = 0.053
[5]Deep learning-based CCA fusion determining RGP lens base-curve on Pentacam four refractive mapsDeep learning feature extraction and CCA feature fusionMSE = 0.056
[7]Deep learning-based autoencoder fusion determining RGP lens base-curve on Pentacam four refractive mapsDeep learning feature extraction and CAE feature fusionMSE = 0.014
R2 = 0.53
[33]Novel deep learning approach to estimate RGP contact lens base-curve for keratoconus fittingUsing the U-net architectureMSE = 0.05
[32]Classification of Pentacam images into four categories: type I to III keratoconus and natural eyesIntegrate multiple classifiers as an expert system and a gateway to combine the results of categories (each category is trained with a specific piece of training data)Accuracy = 92%
[12]Categorization of Pentacam images is categorized into three classes: keratoconus, subclinical keratoconus, and normal cornea with Pentacam imagesPre-processing of images and a convolutional neural networkAccuracy = 98%

CAE: convolutional autoencoder; CCA: canonical correlation analysis; RGP: rigid gas permeable.