Comparison of Ocular Biomechanical Machine Learning Classifiers for Glaucoma Diagnosis

Application of machine learning methodology on patient functional and structural data has been shown to improve glaucoma classification accuracy. Intraocular pressure (IOP) and the biomechanical behaviors of the eye are early indicators of glaucoma. Four classifiers: linear logistic regression, support vector machine, random forest classifier and gradient boosting classifier are tested for discrimination using a 52-patient biomechanical data set that includes 20 glaucoma (40 eyes) and 32 healthy subjects (64 eyes). Results show that the 98.3% accuracy from linear logistic regression (LLR) is the highest correlation accuracy amongst the tested methods. The LLR classification accuracy is comparable with classification accuracies attained for classification using functional and structural measurements from image data sets. Since IOP elevation and biomechanical changes often precede imageable symptoms, the new biomechanics diagnostic classifier maybe used as a detection method for early stage glaucoma.