Diabetes has been one of mankind’s long standing villains. This disease that comes in various forms has been a thorn upon our society for many years now. The amount of issues it can cause in our bodies is second to none if not attended to diligently. Loss of vision is one such issue that can occur due to diabetic macular edema. This condition has been common amongst diabetic working adults making it a cause for major concern. And although there are several methods of treatments in existence, selecting the right procedure depending on the specific condition of the patient has always been a challenge for doctors. For this reason, researchers from the Duke have developed an A.I approach to help doctors take the most appropriate route.
Since some treatment procedures such as the usage of Anti-vascular endothelial growth factor (VEGF) agents involves the application of multiple expensive injections and can be burdensome to all parties involved, it is clear that such procedures must be carefully selected. Moreover, such procedures may not even work for any given patient due to the role of various different factors and must not be considered as a viable option just based on popularity.
But now with the help of an algorithm developed by the researchers at Duke University, doctors can predict whether a patient would respond well to anti-VEGF treatments or not. This is made possible through the algorithm’s ability to automatically analyze optical coherence tomography images (OCT) of the retina. Moreover, the efficiency of the algorithm is also very high as it can make accurate predictions based on just one pre treatment volumetric scan.
This versatile algorithm that can be adapted to cover several other eye related diseases is based on a convolutional neural network (CNN) architecture. By assigning priorities to various aspects or objects this type of artificial intelligence can extensively analyse images. The appropriately named, Predicting treatment response algorithm preserves and highlights global structures in OCT images while making local features from diseased regions clearer. Additionally, the algorithm can also scan for CNN-encoded features that are strongly linked with anti-VEGF response.
To test their algorithm, the researchers fed it with OCT images from 127 different patients who had been treated for diabetic macular edema with three consecutive injections of anti-VEGF agents. Following this, the researchers compared the algorithm’s predictions to images taken before and after anti-VEGF treatments. And based on these results, they found that the algorithm would an 87% chance of correct prediction, with 85% of specificity and 80% sensitivity.
With this approach, patients can potentially avoid unnecessary and expensive trial and error treatments and relieve themselves from substantial physical and cost overheads. Moreover, with the researchers plan to extend the findings from this study to other areas by performing larger observational trials, the eventual adoption of predicting treatment response based techniques will only become imminent.