Purchase this article with an account.
Leon Alexander von der Emde, Maximilian Pfau, Chantal Dysli, Sarah Thiele, Sandrine Künzel, Philipp T. Möller, Matthias Schmid, Monika Fleckenstein, Frank G Holz, Steffen Schmitz-Valckenberg; Longitudinal AI-based assessment of the association between retinal microstructure with rod and cone function in exudative age related macular degeneration. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1202.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To investigate the impact of changes in retinal microstructure on cone and rod function in neovascular age related macular degeneration (nAMD) over time and to evaluate the predictive accuracy of the applied models at follow up examination.
Fifty eyes of 50 nAMD patients (age 76.1 ± 7.6 years) were enrolled for longitudinal multimodal imaging as well as mesopic and two-color dark-adapted (DA) fundus controlled perimetry (FCP, “microperimetry”) over a 12-month period. The inner retina, outer nuclear layer, inner and outer photoreceptor segments and retinal-pigment-epithelium-drusen-complex were semi-automatically annotated in spectral domain optical coherence tomography (SD-OCT) scans (121 B-scans, 30°x25°). FCP data were registered to SD-OCT and fundus autofluorescence data to obtain thickness and intensity data spatially corresponding to the stimulus-location and area (0.43°). Using random forest regression trained with baseline data, we investigated the ability of these multimodal imaging features to predict retinal sensitivity at month 12.
Mesopic sensitivity increased by (mean±SD) 0.49 ± 0.16 dB (p<0.01) over 12 months, while DA cyan sensitivity (-0.89 ± 0.18 dB p<0.01) and DA red sensitivity (-0.78 ± 0.16 dB p<0.01) decreased. Without patient-specific baseline data, the point-wise sensitivity of month 12 could be predicted with a mean absolute error (MAE) of [95% CI] 4.63 dB [3.78, 5.48] for mesopic, 4.42 dB [4.03, 4.81] for DA cyan and 4.61 dB [3.75, 5.47] for DA red testing. Addition of patient specific baseline data improved the MAE for month 12 predictions only minimally. In contrast, partial addition of visit-specific sensitivity data (i.e. 30 of the 61 sensitivity measurements) significantly decreased MAE values with 3.62 dB [3.04, 4.19], 3.91 dB [3.41, 4.4] and 3.76 dB [3.2, 4.32] for the sensitivity predictions of the remaining test-points.
Retinal abnormalities may exhibit differential, yet predictable effects on cone and rod function in nAMD. Partial addition of session-specific FCP data greatly improved the prediction accuracies highlighting the potential advantage of augmenting imaging-based “inferred sensitivity” with brief FCP testing, providing the opportunity to be applied as a functional surrogate endpoint in clinical trials for new therapeutic interventions.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only