A neural network is an artificial intelligence system inspired by the human brain. It comprises a set of units which process information and transfer this between themselves, in a similar manner to brain neurons. While neural networks cannot think like people do, they can undertake much more difficult and complex tasks when studying data, e.g., recognising an image and categorising its visual references.
Aireen uses a cascade of neural networks to assess images. Individual networks are trained in different types of assessment, and together they secure high-quality diagnostics of fundus images, including protecting the patient against operator error (poor image quality, not an eye fundus image).
Assessing quality of images
100 thousand eye fundus images were used to train image-quality models (gradability check). All these images had undergone in-house categorisation, and the outcome was an image-quality assessment model with two categories (gradable and non-gradable)
Detecting diabetic retinopathy
1.2 million images of retinas in various phases of diabetic retinopathy were used to train the diabetic retinopathy detection model. The result of this training is a diabetic retinopathy model which assesses images into two categories: DR or No DR
Used Intel® OpenVINOTM technology makes Aireen® widely accessible and the result is available in seconds.