Glaucoma Diagnostic Accuracy of Machine Learning Classifiers Using Retinal Nerve Fiber Layer and Optic Nerve Data...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on detecting characteristic optic nerve damage.
• Spectral domain optical coherence tomography (SD-OCT) provides quantitative measurements of the retinal nerve fiber layer (RNFL) and optic nerve (ON) parameters.
• Machine learning classifiers (MLCs) offer potential for improved diagnostic accuracy in complex medical conditions.

Purpose of the Study:

• To evaluate the diagnostic performance of MLCs utilizing RNFL and ON parameters from SD-OCT.
• To compare the accuracy of MLCs against established SD-OCT metrics for glaucoma detection.

Main Methods:

• Retrospective analysis of 103 participants (57 glaucoma, 46 healthy).
• Standard ophthalmological examination, visual field testing, and SD-OCT imaging were performed.
• Ten different MLCs were trained and tested using RNFL and ON parameters; diagnostic accuracy was assessed using receiver operating characteristic (ROC) curves and areas under the ROC curve (aROCs).

Main Results:

• The best performing single SD-OCT parameters were cup/disc area ratio (aROC=0.846) and average cup/disc (aROC=0.843).
• MLCs achieved varying aROCs, with the Random Forest (RAN) classifier showing the highest accuracy (aROC=0.877).
• The diagnostic accuracy of the best MLC (RAN) was not statistically significantly different from the best single SD-OCT parameter (P=0.542).

Conclusions:

• MLCs exhibit promising diagnostic accuracy for glaucoma based on SD-OCT data.
• Current MLCs do not offer a significant advantage over traditional SD-OCT parameter analysis for improving glaucoma diagnosis sensitivity and specificity.
• Further research may explore more complex MLC models or integrated multimodal data for enhanced glaucoma detection.