Purchase this article with an account.
Susan Wakil, Alessandro Adad Jammal, Nara Ogata, Felipe Medeiros; A Structural and Functional Machine Learning Classifier Improves Prediction of Patient-Reported Disability in Glaucoma. Invest. Ophthalmol. Vis. Sci. 2018;59(9):4985.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop a machine learning classifier combining structural and functional testing to improve prediction of patient-reported disability in glaucoma.
The study included 487 patients diagnosed with glaucoma. Patient-reported quality of life was assessed by the National Eye Institute Visual Function questionnaire (NEI-VFQ 25). Patients were classified as disabled versus non-disabled based on a latent class analysis of NEI VFQ-25 data. All patients also had testing with standard automated perimetry (SAP SITA 24-2) and spectral-domain optical coherence tomography (SDOCT) in both eyes. Estimates of integrated binocular sensitivity were obtained for each perimetric location. “Binocular” retinal nerve fiber layer (RNFL) measurements were estimated by selecting the better (thicker) measurements between the two eyes for each location. A Random Forest Classifier was trained to discriminate disabled versus non-disabled subjects based on SAP and SDOCT features. An Adaptive Synthetic Sampling Technique (ADASYN) was used to address class imbalance during training. An independent test sample (30% of the original sample) was used to validate the classifier. Classification performance was assessed by the area under the receiver operating characteristic (ROC) curve.
83 of the 487 (17%) patients were classified as disabled based on NEI VFQ-25 results. Mean age was not significantly different between disabled and non-disabled subjects (67.2 ± 12.2 vs. 65.5 ± 12.4; P=0.27). There was a significant difference in average binocular SAP MS between the two groups (27.3 ± 4.2 dB vs 29.3 ± 3.3dB; P<0.001). The area under the ROC curve on the independent test sample was 0.85 (95% CI: 0.81 – 0.90) for the Random Forest Classifier compared to 0.66 (95% CI: 0.60 – 0.73; P<0.001) for SAP MS and 0.62 (95% CI: 0.54 – 0.69; P<0.001) for global RNFL thickness.
A machine learning classifier using structural and functional information significantly improved prediction of patient-reported disability in glaucoma.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
This PDF is available to Subscribers Only