RETFound has shown promise in detecting ocular diseases such as diabetic retinopathy and predicting the risk for systemic diseases. The model is publicly available, and researchers hope it can be adapted and trained to work for different patient populations and medical settings. However, there are concerns that any limitations in RETFound could be passed on to future models built from it, emphasizing the need for ethical and safe usage, and transparent communication of its limitations.
Key takeaways:
- Scientists have developed an AI tool called RETFound that can diagnose and predict the risk of multiple health conditions based on retinal images. The tool was developed using self-supervised learning, eliminating the need for time-consuming and expensive labeling of images.
- RETFound uses a method similar to the one used to train large-language models such as ChatGPT, using a multitude of retinal photos to learn how to predict what missing portions of images should look like.
- The system performed well at detecting ocular diseases such as diabetic retinopathy, and while its performance was limited when predicting the risk for systemic diseases, it was still superior to other AI models.
- The authors have made the model publicly available for adaptation and training for different patient populations and medical settings. However, there are concerns that any limitations in RETFound could be passed on to future models built from it.