IBM and New York University researchers have improved deep learning techniques to better spot glaucoma and detect it early.
Big Blue researchers will present their findings at Association for Research in Vision and Ophthalmology's annual meeting.
In a nutshell, deep learning models can be trained to learn from retina images and then estimate visual function. Those estimates can then be used as a glaucoma indicator. The hope is that non-invasive retina imaging can diagnose glaucoma faster.
Typical visual function tests are based on patient feedback and multiple tests. Glaucoma, the second leading cause of blindness in the world, develops slowly and often diminishes visual function before a diagnosis.
IBM Research and NYU's study used 3D raw Optical Coherence Tomography imaging data to estimate visual field index (VFI) values with an error rate within 2 percent. That error rate was better than tests in the field today.
The VFI is a metric that captures the entire visual field. With artificial intelligence health professionals can estimate visual function and gather data for a glaucoma diagnosis.
IBM and NYU researchers will present 7 abstracts at the annual meeting.
- "Inference of visual field test results from OCT volumes using deep learning"
- "Onesize fits all: OCT image enhancement via deep learning"
- "Deformation Analysis of 3D Optic Cup Surface in Healthy and Glaucoma Patients"
- "Feature agnostic networks outperform classical machine learning approaches in the detection of glaucoma in OCT volumes"
- "Estimating visual field functions in glaucoma patients using multiregional neural networks on OCT images"
- "Forecasting Visual Field parameters at the Future visits using machine learning regression"
- "Deep Learning Based Features Improves Forecasting OCT Measurements at the Future Visit"