2023-09-14 13:30:00
After training RETFound on around 1.6 million unclassified retinal images, Keane and his colleagues were able to insert a small number of classified images – for example, 100 retinal images from people who had Parkinson’s disease and 100 from people who did not – to teach the model certain disease characteristics. After learning what a healthy retina should look like from all the unlabeled images, the model was then able to identify the retinal features associated with disease.
The system performed well in detecting eye diseases such as diabetic retinopathy. On a scale where 0.5 represents a model that predicts no better than chance and 1 represents a perfect model that predicts correctly every time, RETFound for diabetic retinopathy scored between 0.822 depending on the data set used and 0.943. When predicting the risk of systemic diseases such as heart attack, heart failure, stroke and Parkinson’s, overall performance was limited but still better than other AI models.
Can it be extended to other imaging methods?
RETFound is one of the few successful applications of a basic model in medical imaging, says Xiaoxuan Liu, who researches responsible innovation in AI at the University of Birmingham and was not involved in the development of the new process. Now, however, it is a matter of finding out for which other types of medical imaging the techniques used in the development of RETFound might be used – for example on magnetic resonance or computer tomography scans, which are often three or even four dimensional.
The authors have made their model publicly available and hope that groups around the world will be able to adapt and train it for their own patient populations and medical settings. “They can take this algorithm and refine it using data from their own country to optimize it for their purposes,” says Pearse Keane.
1694699247
#detects #eye #diseases #Parkinsons #disease #retinal #images