Machine-learning based forecasting of visual field deficits for glaucoma management

Summary
The current technology is a machine learning algorithm that can accurately forecast glaucoma-induced visual field deficits and outperforms existing forecasting methods. This technology could allow for more efficient management of glaucoma by reducing the number of office-based visual field assessments.

Technology Overview
Glaucoma is characterized by a progressive loss of the visual field (VF). As such, physicians currently rely on frequent office-based testing of the VF to monitoring these patients. Conventional VF trend analysis requires a minimum of three previous VF assessments and provides only a rough estimate of when losses in vision may occur. More recently AI-based methods to predict VF have been developed, but these methods have not been widely adopted as they still lack accuracy.

The current technology is a deep learning-based model that accurately forecasts VFs in glaucoma patients for up to 4 years. Advantages of the technology include:

• Only one VF assessment is needed as input to accurately forecast future glaucoma-induced changes in the VF.
• Flexibility to incorporate multiple VF assessments, which produces even more accurate forecasting for longer periods of time.
• Better performance when compared to conventional trend-based analysis methods and state-of-the-art VF forecasting methods (CascadeNet-5 proposed by Wen et al. and an LSTM based method by Park et al.).

Publications
Ashkan Abbasi, et al., A Unified Framework for Visual Field Test Estimation and Forecasting using Convolution and Attention Networks. Invest. Ophthalmol. Vis. Sci. 2024;65(7):2361. Link

Hiroshi Ishikawa, et al., How Far in the Future Can a Deep Learning Model Forecast Pointwise Visual Field (VF) Data Based Solely on One VF Data Input. Invest. Ophthalmol. Vis. Sci. 2024;65(7):373. Link

Licensing Opportunity
This technology is available for licensing.