Adopting machine learning to automatically identify candidate patients for corneal refractive surgery

Tae Keun Yoo, Ik Hee Ryu, Geunyoung Lee, Youngnam Kim, Jin Kuk Kim, In Sik Lee, Jung Sub Kim, Tyler Hyungtaek Rim, Tae Keun Yoo, Ik Hee Ryu, Geunyoung Lee, Youngnam Kim, Jin Kuk Kim, In Sik Lee, Jung Sub Kim, Tyler Hyungtaek Rim

Abstract

Recently, it has become more important to screen candidates that undergo corneal refractive surgery to prevent complications. Until now, there is still no definitive screening method to confront the possibility of a misdiagnosis. We evaluate the possibilities of machine learning as a clinical decision support to determine the suitability to corneal refractive surgery. A machine learning architecture was built with the aim of identifying candidates combining the large multi-instrument data from patients and clinical decisions of highly experienced experts. Five heterogeneous algorithms were used to predict candidates for surgery. Subsequently, an ensemble classifier was developed to improve the performance. Training (10,561 subjects) and internal validation (2640 subjects) were conducted using subjects who had visited between 2016 and 2017. External validation (5279 subjects) was performed using subjects who had visited in 2018. The best model, i.e., the ensemble classifier, had a high prediction performance with the area under the receiver operating characteristic curves of 0.983 (95% CI, 0.977-0.987) and 0.972 (95% CI, 0.967-0.976) when tested in the internal and external validation set, respectively. The machine learning models were statistically superior to classic methods including the percentage of tissue ablated and the Randleman ectatic score. Our model was able to correctly reclassify a patient with postoperative ectasia as an ectasia-risk group. Machine learning algorithms using a wide range of preoperative information achieved a comparable performance to screen candidates for corneal refractive surgery. An automated machine learning analysis of preoperative data can provide a safe and reliable clinical decision for refractive surgery.

Keywords: Eye manifestations; Machine learning.

Conflict of interest statement

Competing interestsTHR was a scientific advisor to a start-up company called Medi-whale, Inc. GL and YK are employee of Medi-whale, Inc. They received salary or stock as a part of the standard compensation package. The remaining authors declare no competing interests.

Figures

Fig. 1
Fig. 1
Schematic illustrating the purpose of this study
Fig. 2
Fig. 2
Heatmaps representing the predictive performance (AUC) of feature selection and machine learning methods to predict candidates for corneal refractive surgery. This figure shows the results from the tenfold cross-validation procedure. a Support vector machine. b Artificial neural networks. c Random forest. d Least absolute shrinkage and selection operator (LASSO). e AdaBoost
Fig. 3
Fig. 3
The ROC curves for the machine learning algorithms and classic screening methods. a The ROC curves of the internal validation set. b The ROC curves of the external validation set. The machine learning classifiers include random forest (RF), AdaBoost, artificial neural networks (ANN), and ensemble classifier. The classic methods include percentage of tissue ablated (PTA) and Randleman ectatic score
Fig. 4
Fig. 4
The classification performance of high-risk subgroups according to the tenfold cross-validation results. The performance was measured based on the average of the AUCs. The error bars represent the 95% confidence intervals. a Performances in the high myopia group. b Performances in the high astigmatism group. c Performances in the thin corneal thickness group. Error bars indicate the standard deviation of the mean
Fig. 5
Fig. 5
Outcome value histograms of the ensemble machine learning technique in the tenfold cross-validation. The misclassified samples with an opposite outcome value are shown
Fig. 6
Fig. 6
Machine learning technique performance in the ectasia-risk groups, including post-LASIK ectasia, keratoconus, and forme fruste keratoconus patients. a Accuracy in each ectasia-risk group. b ROC curves for classification between the normal control (no postoperative ectasia, N = 9556) and total ectasia-risk group (N = 153)
Fig. 7
Fig. 7
An architecture of our proposed machine learning system to predict candidates for corneal refractive surgery

References

    1. Price MO, et al. Three-year longitudinal survey comparing visual satisfaction with LASIK and contact lenses. Ophthalmology. 2016;123:1659–1666. doi: 10.1016/j.ophtha.2016.04.003.
    1. Yuen LH, et al. A 10-year prospective audit of LASIK outcomes for myopia in 37,932 eyes at a single institution in Asia. Ophthalmology. 2010;117:1236–1244.e1. doi: 10.1016/j.ophtha.2009.10.042.
    1. Alarcón A, Rubiño M, Pééérez-Ocón F, Jiménez JR. Theoretical analysis of the effect of pupil size, initial myopic level, and optical zone on quality of vision after corneal refractive surgery. J. Refract. Surg. Thorofare NJ 1995. 2012;28:901–906.
    1. Díaz JA, Anera RG, Jiménez JR, Del Barco LJ. Optimum corneal asphericity of myopic eyes for refractive surgery. J. Mod. Opt. 2003;50:1903–1915. doi: 10.1080/09500340308235245.
    1. Sayegh FN. Age and refraction in 46,000 patients as a potential predictor of refractive stability after refractive surgery. J. Refract. Surg. Thorofare NJ 1995. 2009;25:747–751.
    1. Salomão M, et al. Recent developments in keratoconus diagnosis. Expert Rev. Ophthalmol. 2018;13:329–341. doi: 10.1080/17469899.2018.1555036.
    1. Caixinha M, Nunes S. Machine learning techniques in clinical vision sciences. Curr. Eye Res. 2017;42:1–15. doi: 10.1080/02713683.2016.1175019.
    1. Rohm M, et al. Predicting visual acuity by using machine learning in patients treated for neovascular age-related macular degeneration. Ophthalmology. 2018;125:1028–1036. doi: 10.1016/j.ophtha.2017.12.034.
    1. Oh E, Yoo TK, Hong S. Artificial neural network approach for differentiating open-angle glaucoma from glaucoma suspect without a visual field test. Invest. Ophthalmol. Vis. Sci. 2015;56:3957–3966. doi: 10.1167/iovs.15-16805.
    1. Kovács I, et al. Accuracy of machine learning classifiers using bilateral data from a Scheimpflug camera for identifying eyes with preclinical signs of keratoconus. J. Cataract Refract. Surg. 2016;42:275–283. doi: 10.1016/j.jcrs.2015.09.020.
    1. Lopes BT, et al. Enhanced tomographic assessment to detect corneal ectasia based on artificial intelligence. Am. J. Ophthalmol. 2018;195:223–232. doi: 10.1016/j.ajo.2018.08.005.
    1. Ambrósio R, et al. Integration of Scheimpflug-based corneal tomography and biomechanical assessments for enhancing ectasia detection. J. Refract. Surg. Thorofare NJ 1995. 2017;33:434–443.
    1. Bower KS, Woreta F. Update on contraindications for laser-assisted in situ keratomileusis and photorefractive keratectomy. Curr. Opin. Ophthalmol. 2014;25:251–257. doi: 10.1097/ICU.0000000000000055.
    1. Ambrósio R, Klyce SD, Wilson SE. Corneal topographic and pachymetric screening of keratorefractive patients. J. Refract. Surg. Thorofare NJ 1995. 2003;19:24–29.
    1. Randleman JB, Trattler WB, Stulting RD. Validation of the Ectasia Risk Score System for preoperative laser in situ keratomileusis screening. Am. J. Ophthalmol. 2008;145:813–818. doi: 10.1016/j.ajo.2007.12.033.
    1. Ramos IC, et al. Variability of subjective classifications of corneal topography maps from LASIK candidates. J. Refract. Surg. Thorofare NJ 1995. 2013;29:770–775.
    1. Hwang Eric S., Perez-Straziota Claudia E., Kim Sang Woo, Santhiago Marcony R., Randleman J. Bradley. Distinguishing Highly Asymmetric Keratoconus Eyes Using Combined Scheimpflug and Spectral-Domain OCT Analysis. Ophthalmology. 2018;125(12):1862–1871. doi: 10.1016/j.ophtha.2018.06.020.
    1. Steinberg J, et al. Screening or keratoconus with new dynamic biomechanical in vivo Scheimpflug analyses. Cornea. 2015;34:1404–1412. doi: 10.1097/ICO.0000000000000598.
    1. Casanova R, et al. Application of random forests methods to diabetic retinopathy classification analyses. PLoS ONE. 2014;9:e98587. doi: 10.1371/journal.pone.0098587.
    1. Mitra J, et al. Lesion segmentation from multimodal MRI using random forest following ischemic stroke. NeuroImage. 2014;98:324–335. doi: 10.1016/j.neuroimage.2014.04.056.
    1. Wei L, Yang Y, Nishikawa RM, Jiang Y. A study on several machine-learning methods for classification of malignant and benign clustered microcalcifications. IEEE Trans. Med. Imaging. 2005;24:371–380. doi: 10.1109/TMI.2004.842457.
    1. Bashir S, Qamar U, Khan FH. Heterogeneous classifiers fusion for dynamic breast cancer diagnosis using weighted vote based ensemble. Qual. Quant. 2015;49:2061–2076. doi: 10.1007/s11135-014-0090-z.
    1. Anera RG, Castro JJ, Jiménez JR, Villa C, Alarcón A. Optical quality and visual discrimination capacity after myopic LASIK with a standard and aspheric ablation profile. J. Refract. Surg. Thorofare NJ 1995. 2011;27:597–601.
    1. Jiménez JR, Alarcón A, Anera RG, Del Barco LJ. Hyperopic Q-optimized algorithms: a theoretical study on factors influencing optical quality. Biomed. Opt. Express. 2017;8:1405–1414. doi: 10.1364/BOE.8.001405.
    1. Jiménez JR, Rodríguez-Marín F, Anera RG, Jiménez Del Barco L. Deviations of Lambert-Beer’s law affect corneal refractive parameters after refractive surgery. Opt. Express. 2006;14:5411–5417. doi: 10.1364/OE.14.005411.
    1. Jiménez JR, Alarcón A, Anera RG, Jiménez Del Barco L. Q-optimized algorithms: theoretical analysis of factors influencing visual quality after myopic corneal refractive surgery. J. Refract. Surg. Thorofare NJ 1995. 2016;32:612–617.
    1. Ambrósio R, Valbon BF, Faria-Correia F, Ramos I, Luz A. Scheimpflug imaging for laser refractive surgery. Curr. Opin. Ophthalmol. 2013;24:310–320. doi: 10.1097/ICU.0b013e3283622a94.
    1. Ambrósio R, Jr, et al. Assessing ectasia susceptibility prior to LASIK: the role of age and residual stromal bed (RSB) in conjunction to Belin-Ambrósio deviation index (BAD-D) Rev. Bras. Oftalmol. 2014;73:75–80. doi: 10.5935/0034-7280.20140018.
    1. Joyaux JC, et al. Comparison of corneal thickness and curvature in myopic and keratoconic eyes of patients from French Caribbean Isles and continental France. Acta Ophthalmol. 2012;90:e529–e533. doi: 10.1111/j.1755-3768.2012.02487.x.
    1. Akay MF. Support vector machines combined with feature selection for breast cancer diagnosis. Expert Syst. Appl. 2009;36:3240–3247. doi: 10.1016/j.eswa.2008.01.009.
    1. Breiman L. Random forests. Mach. Learn. 2001;45:5–32. doi: 10.1023/A:1010933404324.
    1. Chen P, Pan C. Diabetes classification model based on boosting algorithms. BMC Bioinforma. 2018;19:109. doi: 10.1186/s12859-018-2090-9.
    1. Roth V. The generalized LASSO. IEEE Trans. Neural Netw. Publ. IEEE Neural Netw. Counc. 2004;15:16–28. doi: 10.1109/TNN.2003.809398.
    1. Kuncheva LI, Rodríguez JJ. A weighted voting framework for classifiers ensembles. Knowl. Inf. Syst. 2014;38:259–275. doi: 10.1007/s10115-012-0586-6.
    1. Liang D, Tsai C-F, Wu H-T. The effect of feature selection on financial distress prediction. Knowl. -Based Syst. 2015;73:289–297. doi: 10.1016/j.knosys.2014.10.010.
    1. Chen Y, Wang S, Shen C-H, Choy FK. Matrix decomposition based feature extraction for murmur classification. Med. Eng. Phys. 2012;34:756–761. doi: 10.1016/j.medengphy.2011.09.020.
    1. Hall MA, Holmes G. Benchmarking attribute selection techniques for discrete class data mining. IEEE Trans. Knowl. Data Eng. 2003;15:1437–1447. doi: 10.1109/TKDE.2003.1245283.
    1. Kononenko, I. Estimating attributes: analysis and extensions of RELIEF. in European Conference on Machine Learning. (eds Bergadano, F. & Raedt, L. D.) 171–182 (Springer, New York, 1994).
    1. Wu W, et al. Exploratory study to identify radiomics classifiers for lung cancer histology. Front. Oncol. 2016;6:71.
    1. Dudoit S, Fridlyand J, Speed TP. Comparison of discrimination methods for the classification of tumors using gene expression data. J. Am. Stat. Assoc. 2002;97:77–87. doi: 10.1198/016214502753479248.
    1. Randleman JB, Russell B, Ward MA, Thompson KP, Stulting RD. Risk factors and prognosis for corneal ectasia after LASIK. Ophthalmology. 2003;110:267–275. doi: 10.1016/S0161-6420(02)01727-X.
    1. Santhiago MR, et al. Association between the percent tissue altered and post-laser in situ keratomileusis ectasia in eyes with normal preoperative topography. Am. J. Ophthalmol. 2014;158:87–95.e1. doi: 10.1016/j.ajo.2014.04.002.
    1. Chan C, et al. Analysis of cases and accuracy of 3 risk scoring systems in predicting ectasia after laser in situ keratomileusis. J. Cataract Refract. Surg. 2018;44:979–992. doi: 10.1016/j.jcrs.2018.05.013.
    1. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–845. doi: 10.2307/2531595.
    1. Bradley AP. The use of the area under the ROC curve in the evaluation of machine learning algorithms. Pattern Recognit. 1997;30:1145–1159. doi: 10.1016/S0031-3203(96)00142-2.

Source: PubMed

스폰서 및 공동 작업자

건강 상태

약물 개입

3
구독하다