Researchers found that using automated methods to determine the VA showed “robust performance in predicting visual impairment”
Researchers from the US and the Netherlands collaborated in an effort to determine if using confocal scanning laser ophthalmoscopy (cSLO) imaging and deep learning (DL) can estimate the visual acuity (VA) in patients with retinitis pigmentosa (RP). They found that using automated methods to determine the VA showed “robust performance in predicting visual impairment” in this patient population,1 according to Tin Yan Alvin Liu, MD, the first author from the Wilmer Eye Institute, Johns Hopkins Hospital, Baltimore.
The investigators explained that the clinical trials for treatments for RP are limited because of the screening burden and the absence of reliable surrogate markers for functional endpoints. They conducted the study under discussion because of the belief that automated methods to determine the VA may address such challenges.
The Snellen corrected VA and cSLO images were obtained retrospectively from the Johns Hopkins University and the Amsterdam University Medical Centers, both of which used the same exclusion criteria: visually significant media opacities and images not centered on the central macula. The Johns Hopkins dataset were used for 10-fold cross-validations and internal testing; the Amsterdam University Medical Centers dataset was used for external independent testing. The Johns Hopkins data included patients with RP with/without molecular confirmation, and that from the Netherlands included only molecularly confirmed patients with RP. Using transfer learning, they explained, 3 versions of the ResNet-152 neural network were trained: infrared (IR), optical coherence tomography (OCT), and combined image (CI).
Regarding internal testing, the results reflected the findings of 2,569 images obtained from 462 eyes of 231 patients; the area under the curve (AUC) values for distinguishing between Snellen VA 20/40 or better and worse than Snellen VA 20/40 were 0.83, 0.87, and 0.85 for IR, OCT, and CI, respectively. Regarding internal testing, the results reflected the findings from 349 images obtained from 166 eyes of 83 patients; the respective values for IR, OCT, and CI were 0.78, 0.87, and 0.85.
The investigators concluded that the algorithm showed robust performance in predicting visual impairment in patients with RP, thus providing proof-of-concept for predicting a structure-function correlation based solely on cSLO imaging in patients with RP. This deep learning-based estimation of VA using OCT images may enable efficient screening of potential subjects in future RP research studies or clinical trials.