Deep-Learning System Can Feasibly Detect Retinal Lesions

Retinal Detachment
Large Rip And Detachment Of Retina. (Photo By BSIP/UIG Via Getty Images)
Researchers train artificial intelligence to recognize retinal breaks and detachment in tessellated eyes.

Using ultrawide-field images in a deep-learning system works to detect lattice degeneration, retinal breaks, and retinal detachment in tessellated eyes, according to a study published in Graefe’s Archive for Clinical and Experimental Ophthalmology.

In addition, the system might work for screening and telemedicine needs, researchers report. It also can be used in patients who cannot tolerate pupil dilation.

The pilot study was designed to investigate 3 retinal lesions in tessellated eyes with an ultrawide-field fundus imaging system using convolutional neural network technology. The 1500 color images were evaluated by 3 retinal investigators for tessellated fundus confirmation and assessment of peripheral retinal lesion (lattice degeneration, retinal breaks, and retinal detachment).

Of those images, 722 were used to train and verify the combined deep-learning system. A test set of 189 images verified performance, comparing it to the reference standard.

Researchers found that referral accuracy of the system was 79.8% compared with the reference standard.

“With optimal preprocessing approach (original resizing method for lattice degeneration and retinal detachment, cropping method for retinal breaks), the combined deep-learning system exhibited an area under curve of 0.888, 0.953, and 1.000 for detection of lattice degeneration, retinal breaks, and retinal detachment respectively in tessellated eyes,” the study says.

The study’s  limitations include its pilot study design, relative small dataset, the fact it was not validated by an independent external dataset, only 3 target retinal lesions were included, and unclear images from vitreous hemorrhage or cataracts were excluded. And while the ultrawide-field imaging technology can capture up to 200 °, peripheral regions — especially those superior and inferior peripheral areas hidden by eyelids and eyelashes — are not covered, according to the report.


Zhang C, He F, Li B, et al. Development of a deep-learning system for detection of lattice degeneration, retinal breaks, and retinal detachment in tessellated eyes using ultra-wide-field fundus images: a pilot study. Graefes Arch Clin Exp Ophthalmol. Published online February 4, 2021. doi:10.1007/s00417-021-05105-3