Purchase this article with an account.
Sidharth Mahotra, Xiaoqin Huang, Tobias Elze, Mengyu Wang, Michael V Boland, Louis Pasquale, Juleke Eugenie Anne Majoor, Koenraad Arndt Vermeer, Hans Lemij, Kouros Nouri-Mahdavi, Chris A Johnson, Siamak Yousefi; Detecting Glaucoma Progression using Deep Archetypal Analysis of Retinal Nerve Fiber Layer Thickness Measurements. Invest. Ophthalmol. Vis. Sci. 2021;62(8):3357.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop an artificial intelligence framework for detecting glaucoma progression using retinal nerve fiber layer (RNFL) thickness measurements obtained from optical coherence tomography (OCT) imaging of the optic disc.
We developed a deep archetypal analysis (DAA) framework and applied it to RNFL thickness measurements (circle scans with 768 A-scans) of 691 eyes of 691 patients and identified 16 prevalent patterns of RNFL loss. We then developed a framework to detect glaucoma progression using the deep archetypes discovered. We simulated a stable dataset by randomly shuffling the longitudinal visits of another dataset with 254 eyes of 127 subjects (Average of 9 visits). We selected the critical slopes of no-progression from this dataset at 95th percentile and subsequently used an independent longitudinal dataset with 254 eyes (mean 9 visits) to compare the detection rate of the proposed model against linear regression of RNFL summary parameters.
Deep archetypal analysis discovered 16 patterns of RNFL loss which explained over 70% of the total variation in the RNFL data (Fig. 1). The critical slopes were selected in such a way to maintain 95% specificity for each model using the simulated stable dataset. The sensitivity of the overall DAA model was 54.7% while the detection rate of the linear regression of RNFL in global, superior, and inferior hemifields were 16.1%,13.0% and 16.9%, respectively. Most of the eyes progressed across deep archetypes number 10 and 12 (Fig. 2).
The proposed deep archetypal analysis framework identifies major patterns of RNFL loss in patients with glaucoma. It also provides a more sensitive model in detecting glaucoma progression compared to linear regression of the summary parameters. This model may aid clinicians in detecting structural progression and identifying the corresponding pattern of RNFL loss, which in turn could improve disease monitoring and treatment planning.
This is a 2021 ARVO Annual Meeting abstract.
Figure 1. Patterns of RNFL loss identified by deep archetypal analysis (DAA). Colors denote RNFL thickness deviation from a cohort of normal eyes in micron unit.
Figure 2. The contribution of each deep archetype in detecting glaucoma progression.
This PDF is available to Subscribers Only