Deep learning model might help monitor progression of retinitis pigmentosa
Key takeaways:
- A deep learning model estimated visual function in patients with retinitis pigmentosa from ultra-widefield fundus autofluorescence images.
- This method could assist clinicians in monitoring RP progression.
A deep learning model showed the ability to estimate visual function in patients with retinitis pigmentosa from ultra-widefield fundus autofluorescence images, which may provide a new way of objectively assessing disease progression.
In a retrospective study involving five institutions in Japan, three types of images — ultra-widefield pseudocolor (UWPC) images, ultra-widefield fundus autofluorescence (UWFAF) images and both UWPC and UWFAF images — obtained from 1,274 eyes of 695 consecutive patients with retinitis pigmentosa (RP) were used to train, validate and test 31 types of ensemble models constructed from five deep learning (DL) models.
Image: Adobe Stock
The model using UWFAF images alone provided the best estimation accuracy for visual acuity, mean deviation on the Humphrey field analyzer and central retinal sensitivity.
“The DL model had higher estimation accuracy from UWFAF images alone likely because this type of image had more information for the model. Thus, the information on the retinal pigment epithelium function reflected in the FAF images could be highly beneficial in estimating visual functions,” the authors wrote.
New treatments for RP are in development, but the current practice mainly involves observation, care for residual visual function and management of complications.
UWFAF images can be obtained easily, noninvasively and without mydriasis, so a DL model that can accurately monitor the visual function from these images in RP patients “might help clinicians assess the progression of RP objectively,” the authors said.
Collapse
Read more about