Abstract
Purpose::
To investigate the diagnostic performance of artificial neural network (ANN) on glaucoma disease in Taiwan Chinese population based on the retinal nerve fiber layer thickness measurement data from scanning laser polarimetry-variable corneal compensation (GDx VCC).
Methods::
This study included on eye each from 54 glaucoma and 54 healthy subjects. Each patient received complete ophthalmological evaluation, an achromatic automated perimetry (AAP), and GDx-VCC exam. All measured GDx VCC parameters were compared between the two groups. Area under receiver operating characteristics (AROC) curve and sensitivities at predetermined specificities of >or=80% and >or=95% for each single parameter were calculated. Based on the data base of Taiwan Chinese group, artificial neural network (ANN) was applied to generate a novel classification index to discriminate between glaucomatous and normal eyes in terms of AROC curve.
Results::
All GDX VCC measured parameters were significantly different between the two groups (p<0.001) except two parameters (the symmetry and ellipse modulation). The three parameters with best AROC curves for differentiating normal from glaucomatous eye were NFI (0.82), superior average thickness (0.806) and inferior average thickness (0.782). The AROC from ANN was 0.984. Compared with NFI, ANN method improved the ability to differentiate between normal and glaucomatous eye in the Taiwan Chinese population from the summary data of RNFL thickness measurements by GDX VCC.
Conclusions::
Results from ANN trained on GDX VCC RNFL thickness measurements show better discrimination power in glaucoma disease. Furthermore, to establish the normal database of each race is mandatory in glaucoma imaging machine, especially to GDX VCC.
Keywords: nerve fiber layer • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)