A Fused Pattern Recognition Model to Detect Glaucoma Using Retinal Nerve Fiber Layer Thickness Measurements
It is estimated that approximately 1.3 billion people live with some form of vision impairment. Glaucomatous optic neuropathy is listed as the fourth major cause of vision impairment by the WHO. In 2015, an estimated 3 million people were blind due to this disease.
Structural and functional methods are utilized to detect and monitor glaucomatous damage. The relationship between these detection measures is complex and differs between individuals, especially in early glaucoma.
In this study, we aim at evaluating the relationship between retinal nerve fibre layer (RNFL) thickness and glaucoma patients. Thus, we develop a fused pattern recognition model to detect healthy vs. glaucoma patients. We also achieved an F1 score of 0.82 and accuracy of 82% using 5-fold cross-validation on a data set of 107 RNFL data from healthy eyes and 68 RNFL data from eyes with glaucoma; 25% of data have been selected randomly for testing.
The proposed fused model is based on a stack of supervised classifiers combined by an ensemble learning method to achieve a robust and generalised model for glaucoma detection in the early stages. Additionally, we implemented an unsupervised model based on K-means clustering with 80% accuracy for glaucoma screening. In this research, we have followed two purposes; first, to assist the ophthalmologists in their daily Patient examination to confirm their diagnosis, thereby increasing the accuracy of diagnosis. The second usage is glaucoma screening by optometrists in order to perform more eye tests and better glaucoma diagnosis.
Therefore, our experimental tests illustrate that having only one data set still allows us to obtain highly accurate results by applying both supervised and unsupervised models. In future, the developed model will be retested on more substantial and diverse data sets.
KeywordsGlaucoma detection RNFL Machine learning Pattern recognition Unsupervised classifier Hybrid classifier Ensemble learning
- 3.Gour, N., Khanna, P.: Automated glaucoma detection using GIST and pyramid histogram of oriented gradients (PHOG) descriptors. Pattern Recogn. Lett. (2019)Google Scholar
- 9.Norouzifard, M., Nemati, A., Klette, R., GholamHosseini, H., Nouri-Mahdavi, K., Yousefi, S.: A hybrid machine learning model to detect glaucoma using retinal nerve fiber layer thickness measurements. Invest. Ophthalmol. Vis. Sci. 60(9), 3924 (2019)Google Scholar
- 10.Palacio-Niño, J.O.: Evaluation metrics for unsupervised learning algorithms. arXiv preprint. arXiv:1905.05667 (2019)
- 13.Sokolova, M., Japkowicz, N., Szpakowicz, S.: Beyond accuracy, F-score and ROC: a family of discriminant measures for performance evaluation. In: Sattar, A., Kang, B. (eds.) AI 2006. LNCS (LNAI), vol. 4304, pp. 1015–1021. Springer, Heidelberg (2006). https://doi.org/10.1007/11941439_114CrossRefGoogle Scholar
- 15.Tharwat, A.: Classification assessment methods. Appl. Comput. Inf. (2018)Google Scholar
- 16.Glaucoma in 2015 (2015). www.iapb.org/knowledge/what-is-avoidable-blindness/glaucoma/
- 17.New Zealand: Latest stats at a glance (2015). www.blindfoundation.org.nz
- 18.World Health Organization (2018). www.who.int/news-room/fact-sheets/detail/blindness-and-visual-impairment. Accessed 22 July 2019