DRUNET (program for assessment of damages of an optic nerve)

2018: The announcement of software for assessment of damages of an optic nerve at glaucoma

In July, 2018 researchers from the National University of Singapore submitted the new software using artificial intelligence for assessment of damages of an optic nerve at glaucoma. The main complication of glaucoma – total loss of sight owing to damage of an optic nerve, however early identification and treatment can delay progressing of a disease.

The modern methods of deep learning applied in an optical coherent tomography can automatically reveal pathological changes at glaucoma, however it is required to apply the specific algorithm to each type of fabric – such approach requires considerable computing powers and is subject to errors.

Researchers from the National University of Singapore provided new approach to deep learning of AI for assessment of damages of an optic nerve at glaucoma

New approach is based on deep learning with only one algorithm which automatically segments and selects six different structural parameters of an optic nerve at the same time. The technology called Dilated-Residual U-Net, or DRUNET U-Net - the neuronet developed for recognition of biomedical images is inspired. The research DRUNET was conducted among 100 people: 41 patients with open-angle glaucoma, 19 with closed-angle glaucoma and 40 healthy volunteers for control. When testing DRUNET segmented tomographic images of an optic nerve, than other technologies of deep learning much better, and revealed almost all local and contextual features of fabrics. At this DRUNET worked quicker as it needed to estimate less than the parameters: only 40,000 in comparison from 140,000 parameters of the previous systems.

Authors of a research recognized that an inspection was carried out with some restrictions: accuracy of an algorithm was estimated on the manual segmentation which is carried out by only one expert observer, and training of a system was provided on the basis of the tomographic images received from one device; so far it is unknown whether DRUNET will be also effective during the work with other devices of an optical coherent tomography.

Researchers hope to expand use of DRUNET for segmentation 3D - images and plan that their development will begin to be used widely in medical institutions in 2018.^[1]

Notes