Diabetes is now a disease which affects c425 million people globally, according to the International Diabetes Federation. Close to half of those with diabetes will experience retinopathy, a result of which can be blindness if not detected at an early stage.
Aladdin has developed a prototype which uses an artificial intelligence-based grading algorithm to detect, with high reliability and accuracy, referable diabetic retinopathy.
Using 37,266 retinal images, of which 80% were used to train the Deep Neural Networks Algorithms and 20% were used to validate the performance, Aladdin’s AI/ML team were able to teach the prototype to identify diabetes patients from healthy patients with the following high reliability:
Accuracy to distinguish Normal/Diabetic Retinopathy: 94.3%
Accuracy to distinguish Non-Proliferative/Proliferative: 96.7%
Furthermore, the prototype can distinguish the different stages of diabetes (None, Mild, Moderate, Severe, Proliferative) to 84.1% accuracy. These are exciting results at such an early stage of development. Aladdin’s AI/ML team will now work to further the accuracy and reliability of the prototype using the increasing amounts of data on Aladdin’s platform.
This development is particularly pertinent as the US Food and Drug Administration (FDA) approved, in April this year, the first software powered by artificial intelligence that replaces the need for a specialized doctor to interpret medical imagery. Validating the assumption that there is significant demand for these technologies within the healthcare industry.
Wade Menpes Smith
End of Media Release