Simulating glaucoma progression in the retinal ganglion cell layer with generative adversarial networks

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a progressive optic neuropathy leading to vision loss.
• Accurate simulation of glaucoma progression is crucial for research and treatment development.
• Current imaging methods may not fully capture the dynamic changes in the ganglion cell layer (GCL).

Purpose of the Study:

• To develop a generative adversarial network (GAN) tool for synthesizing GCL images.
• To simulate glaucoma progression using artificial intelligence.
• To create realistic GCL images for research purposes.

Main Methods:

• Utilized a dataset of 406 GCL images from 76 glaucomatous eyes obtained via spectral domain optical coherence tomography (OCT).
• Employed the Pix2Pix conditional GAN model to generate images representing glaucoma progression.
• Validated image synthesis using structural similarity coefficient and expert evaluation.

Main Results:

• Synthesized images effectively replicated glaucoma lesion patterns with good generalizability.
• Achieved a mean structural similarity coefficient between 0.76 and 0.78.
• Expert evaluation showed 57% accuracy in distinguishing real from generated images.

Conclusions:

• The developed system successfully generates synthetic GCL images with high similarity to real ones.
• The GAN model demonstrates effectiveness in synthesizing images representing glaucoma evolution.
• This tool shows promise for advancing glaucoma research and understanding disease progression.