Abstract
Purpose :
(1) To develop a deep learning algorithm to automatically and simultaneously identify major tissue structures of the optic nerve head (ONH) in 3D optical coherence tomography (OCT) scans; (2) to exploit such information to robustly differentiate among healthy, optic disc drusen (ODD), and papilledema ONHs.
Methods :
This was a cross-sectional comparative study including ethnically diverse patients from 3 sites (in Singapore, Australia, Denmark) with confirmed ODDs (105 eyes), papilledema due to high intracranial pressure (51 eyes), and healthy controls (100 eyes). Volume raster scans of the ONHs were acquired using Spectralis OCT, then processed with adaptive compensation to improve deep-tissue visibility. In a first step, a deep learning algorithm was developed using 984 B-scans (from 130 eyes) in order identify: major neural and connective tissues, and ODD regions whenever present. The performance of our segmentation algorithm was assessed (against manual segmentations) using the dice coefficient. In a second step, a classification algorithm (random forest) was designed using 150 OCT volumes to perform 3-class classifications (class 1: ODD, class 2: papilledema, class 3: healthy) strictly from their drusen and prelamina swelling scores that were directly calculated from the segmentations. To assess performance, we reported the area under the receiver operating characteristic curves (AUCs) for each class (one-vs-all).
Results :
Our segmentation algorithm was able to simultaneously isolate neural & connective tissues, and ODD regions whenever present (Figure). This was confirmed by an averaged Dice coefficient of 0.93±0.03 on the test set, corresponding to very good segmentation performance. Classification was achieved with very high AUCs, i.e. 0.99±0.01 for the detection of ODD, 0.99±0.01 for the detection of papilledema, and 0.98±0.02 for the detection of healthy ONHs.
Conclusions :
A relatively simple AI approach allows to accurately discriminate ODD from papilledema, strictly using a single OCT scan of the ONH. Our classification performance was excellent, with the caveat that validation in a much larger population is warranted for clinical acceptance. Our approach may have the potential to establish OCT as the mainstay of diagnostic imaging for optic nerve disorders in neuro-ophthalmology.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.