It seems that now, AI is able to detect ocular disease. Check out the latest reports about the matter below.
Recent advancements in AI technology have resulted in highly accurate identification of eye diseases in retinal images, improving diagnostic accuracy and risk assessment. These models incorporate both natural images and medical data to provide reliable disease predictions, allowing for efficient risk stratification in fields such as chest X-rays and dermatology imaging.
A recent study published in the journal Nature highlights RETFound, a foundation model for retinal images that utilizes self-supervised learning (SSL) and a masked autoencoder. By learning useful representations from unlabeled retinal images, RETFound can be adapted for various applications with fewer labeled images required.
Breakthrough study is out
The RETfound study investigated the capability of diagnosing ocular illnesses, predicting their prognosis, and identifying ocular problems. Additionally, it estimated the risk of cardiovascular disorders, including myocardial infarction, cardiac failure, and ischemic stroke, as well as neurodegenerative diseases such as Parkinson’s disease over a three-year period.
The study evaluated the ability of RETfound to detect illnesses via controlled tests and qualitative analysis and assessed its performance and adaptability to various ocular activities after being trained on retinal scans.
Study conclusions
The model, the Adapted RETFound, has shown exceptional performance in identifying eye diseases that can cause vision loss, forecasting the development of serious systemic illnesses such as heart failure and myocardial infarction, and predicting the course of illnesses, despite having limited labeled data. RETFound consistently outperformed other cutting-edge models, including those trained on ImageNet-21k using traditional transfer learning, in terms of both performance and label efficiency.
Based on the findings from ocular and oculomic research, the areas that were most notable were consistent with prior knowledge. RETFound showed the highest level of performance among the different datasets, followed by SL-ImageNet.
To conclude, RETFound is a trustworthy approach for improving retinal imaging and enhancing the diagnostic and prognostic capabilities of AI applications.
