RE: Multicenter reliability of semiautomatic retinal layer segmentation using OCT
JagannadhaAvasarala, Neurologist and neuroimmunologist, Greenville Health System
Submitted April 23, 2018
This article by Oberwahrenbrock et al.[1] is for a highly selective audience who dabble in OCT. Almost 99.9% of the neurology audience will not follow such intricate details that the authors have so aptly laid out. This is because most university neurologists and neuroimmunologists do not even use fundus camera images to educate students and capture data of MS patients in the OP setting, let alone use OCTs to understand the nitty gritty of retinal segmentation using 'standardized protocols', of which there are none . In addition, this study suffers from a strong selection bias given the nature of this topic. It almost sounds as if the semiautomatic retinal layer segmentation algorithms sold by Spectralis are being validated. However, what of the centers that use Cirrus OCT? How will they 'normalize' their data in large clinical trials that need their input as well the experts that reside in those centers? Moreover, in the ideal world, retinal layer segmentation by experts ought to be confirmed by histological data and not algorithm-driven data analysis. In a recent study,[2] the segmentation of retinal layers among 3 OCT angiography instruments in the central macula was shown to be incorrect because of inherent errors in segmentation; studies of the superficial and deep vascular plexuses using manufacturer-recommended default settings were found to be biased. This finding was brought to light owing to the fact that histological data were used as the 'ground truth'.
References
1. Oberwahrenbrock T, Traber GJL, Lukas S, et al. Multicenter reliability of semiautomatic retinal layer segmentation using OCT. Neurol Neuroimmunol Neuroinflamm 2018; 5:e449.
2. Spaide RF, Curcio CA. Evaluation of segmentation of the superficial and deep vascular layers of the retina by optical coherence tomography angiography instruments in normal eyes. JAMA Ophthalmol. 2017;135:259-262.
This article by Oberwahrenbrock et al.[1] is for a highly selective audience who dabble in OCT. Almost 99.9% of the neurology audience will not follow such intricate details that the authors have so aptly laid out. This is because most university neurologists and neuroimmunologists do not even use fundus camera images to educate students and capture data of MS patients in the OP setting, let alone use OCTs to understand the nitty gritty of retinal segmentation using 'standardized protocols', of which there are none . In addition, this study suffers from a strong selection bias given the nature of this topic. It almost sounds as if the semiautomatic retinal layer segmentation algorithms sold by Spectralis are being validated. However, what of the centers that use Cirrus OCT? How will they 'normalize' their data in large clinical trials that need their input as well the experts that reside in those centers? Moreover, in the ideal world, retinal layer segmentation by experts ought to be confirmed by histological data and not algorithm-driven data analysis. In a recent study,[2] the segmentation of retinal layers among 3 OCT angiography instruments in the central macula was shown to be incorrect because of inherent errors in segmentation; studies of the superficial and deep vascular plexuses using manufacturer-recommended default settings were found to be biased. This finding was brought to light owing to the fact that histological data were used as the 'ground truth'.
References
1. Oberwahrenbrock T, Traber GJL, Lukas S, et al. Multicenter reliability of semiautomatic retinal layer segmentation using OCT. Neurol Neuroimmunol Neuroinflamm 2018; 5:e449.
2. Spaide RF, Curcio CA. Evaluation of segmentation of the superficial and deep vascular layers of the retina by optical coherence tomography angiography instruments in normal eyes. JAMA Ophthalmol. 2017;135:259-262.