According to reports, Google said on Tuesday that it has trained a deep learning algorithm that can help detect signs of retinopathy in diabetic patients. The working principle is similar to that of an ophthalmologist who finds signs of the lesion by examining the fundus.
Google teamed up with a team of doctors in India and the United States to create a database of 128,000 images and use the database to train a deep neural network to discover diabetic retinopathy. Related research was published in the Journal of the American Medical Association on Tuesday.
Google said that in the future, this algorithm will help patients get doctors faster, and the algorithm needs further optimization to enter the large-scale promotion stage.
Dr. Lily Peng said in the blog post that "automated, high-precision screening methods may help doctors evaluate more patients and allow patients in need to receive expert treatment more quickly. We hope that this research will become Prove that machine learning helps solve one of the many examples of important medical problems.
According to the National Institutes of Health, diabetic retinopathy makes diabetic patients more painful and blind. It affects blood vessels in the retina and is the most common cause of vision loss in diabetic patients.
This is not the first time Google has been involved in the field of ophthalmology. In 2014, Google and Swiss pharmaceutical company Novartis jointly developed smart contact lenses.
Among them, Google's artificial intelligence (AI) department DeepMind developed "DNC" (differenTIable neural computer) through Google's latest mechanical learning algorithm. Even without prior knowledge, DNC can solve small problems. For example, planning the best route between subway stations, or understanding the family tree relationship.
According to research, the neural network is an interconnected system that mimics the operation of biological neural networks, such as the brain. In recent advances in AI, neural networks have played a key role. Neural networks can infer patterns, for example, it can enhance speech recognition capabilities in digital assistants (Google Voice, Siri).
It is understood that the neural network can only connect to the data contained in its own network. In Nature, the DeepMind team stated that with the support of DNC, the neural network can access previously incompatible external data, such as text encoded in traditional digital mode.
Alex Graves, director of the DNC project, explained the problem of neural networks. He said that the memory of the neural network is constrained by the calculation itself. As a result, the neural network is very fragile and difficult to scale. This problem will make it more powerful by separating the memory so that it can be scaled up without affecting the processor.
Guangzhou Winson Information Technology Co., Ltd. , https://www.barcodescanner-2d.com