Through the combination of fundus cameras, artificial intelligence algorithms, NVIDIA GPUs and edge computing architecture, Taiwanese medical technology company MiiS has applied artificial intelligence to the forefront to combat blindness caused by diabetes .
There are currently more than 400 million diabetic patients in the world, Diabetic eye disease screening has therefore become more and more common, but eye disease screening is not easy.
The diagnosis of symptoms of diabetic retinopathy is a common clinical problem, and it is often difficult to obtain results even with the help of high-resolution images of the inside of the eye. Given that many hospitals and clinics do not have strong Internet connections, it is almost impossible for primary medical staff to use cloud platforms to analyze images and provide comprehensive diagnoses. Even if it is possible, the entire process is very time-consuming.
Julie Chen, vice president of sales at MiiS, a Taiwanese medical device manufacturer, said, "The long diagnosis time increases the possibility that patients miss early treatment opportunities."
MiiS, a Taiwanese medical equipment manufacturer. The company solves this problem through a combination of fundus cameras, artificial intelligence algorithms, NVIDIA GPUs, and edge computing architecture to quickly get diabetic retinopathy diagnoses to doctors.
MiiS essentially gives doctors a A highly portable, GPU-powered artificial intelligence device that generates diagnostic results in seconds.
Julie Chen said: "Patients may not be able to see an ophthalmologist for various reasons. Our smart devices can prevent people from waiting for a long time and delay diagnosis."
There is no doubt: early diagnosis is crucial
Detecting diabetes at an early stage Retinopathy can delay or even prevent blindness caused by diabetes. However, key medical personnel often cannot capture some fuzzy indicators like ophthalmologists or MiiS’s AI model.
To train this model, MiiS obtained a data set containing 120,000 images from three partner hospitals. The images were labeled by a team of 50 doctors and trained on an array of NVIDIA GPUs in the cloud.
Jinhong Technology’s (MiiS) solution first generates the required fundus images, which are then immediately applied to the AI model and run inference calculations on the embedded Jetson TX2. The doctor will receive the analysis results within 5 seconds.
Jinhong Technology’s (MiiS) AI model’s prediction accuracy is approximately 90%, which is comparable to the judgment accuracy of professional ophthalmologists. But the absolute advantage of this solution is speed: the edge computing approach performs 10 times better than cloud-based competitors, and the latter can only run when the enterprise's Internet connection is stable.
Julie Chen said: "General practitioners can know the diagnosis results immediately, and no longer have to wait for feedback from the cloud server ."
The future is promising: this is just the beginning
Although Jinhong Technology (MiiS) was founded in 2010 It has developed a number of diagnostic devices utilizing its Horus series of digital oscilloscopes, but this solution for diabetic retinopathy is its first attempt to combine fundus imaging, AI and edge computing. This solution is currently under FDA review.
Julie Chen said that this method can also be applied to other fields related to diabetes, such as otology, dentistry and dermatology.
In addition, she said the company hopes its application of edge computing will become popular throughout the field of artificial intelligence in the future, whether it is medical or industrial.
Jinhong Technology (MiiS) is a member of the NVIDIA Start-up Program, a virtual accelerator program that provides market promotion, expertise and technical support to start-ups.
