Shulou(Shulou.com)06/02 Report--

Produced by big data Digest

Author: Jiang Baoshang

DeepMind has created a device that can diagnose complex eye diseases in real time, according to the Financial Times. Although the product is only a prototype, it is an important step in Deepmind's strategy of laying out medical devices as a company owned by Alphabet.

At the 2019 Wired Health Conference (Wired Health 2019 conference) last Tuesday, DeepMind scanned the retina of a patient and diagnosed the woman's eyes in real time. The scan results were analyzed by a set of algorithms in Google Cloud that gave emergency ratings and detailed diagnoses in about 30 seconds.

The system can detect a range of eye diseases, including glaucoma, diabetic retinopathy and age-related macular degeneration, with accuracy comparable to that of the world's leading experts.

DeepMind algorithm creates medical products for the first time

In fact, academic research on this product was published in Nature Medicine, a subsidiary of Nature, as early as last August.

According to the paper, this algorithm can identify more than 50 eye diseases that threaten vision by scanning eye OCT images, with an accuracy of 94%, which exceeds the performance of human experts. This is also the first time DeepMind has created a medical product using AI and machine learning algorithms.

Paper address:

Https://www.nature.com/articles/s41591-018-0107-6

Specifically, two independent neural networks are used, a loose machine learning mechanism based on how the human brain works. One neural network marks the features in OCT images related to ophthalmopathy, and the other neural network makes a diagnosis of eye conditions based on these features.

In tests on scanned images of 997 patients, DeepMind's algorithm performed better in referral accuracy than 8 retinal specialists in Moorfields Eye Hospital: the error rate of DeepMind algorithm was 5.5%, while that of 8 human doctors was between 6.7% and 24.1%. If human doctors are provided with background information about patients, the error rate of human doctors is reduced to between 5.5% and 13.1%, which is equal to or slightly worse than the level of AI.

At the time, DeepMind and its research partners, London's Moorfields Eye Hospital and London University College of Ophthalmology, said they would enter clinical trials of the technology in 2019.

Breaking the black box, DeepMind AI algorithm creates medical products for the first time

After months of hard work, DeepMind and Moorfields's team developed the product, but the product has not yet been approved by regulators.

However, this product is likely to be approved by the relevant departments soon. In the United States, the number of medical artificial intelligence algorithms approved by the Food and Drug Administration (FDA) has increased significantly, from two in 2017 to one or two per month in 2018. One of them is an algorithm from IDx, an Iowa-based startup, which, like DeepMind, can scan and diagnose diabetic retinopathy in real time and has been used in clinical care across Europe.

For DeepMind's product, Alan Karthikesalingam, a senior clinician and project leader, said. The birth of this research is an important milestone in the field of medicine, with the help of artificial intelligence will greatly improve the clinical skills of general doctors.

The algorithm is trained in 14884 retinal scans and processes eye images in 30 seconds. The algorithm divides the scan results into four categories: urgent, semi-urgent, conventional and observation-only.

Allen said the goal of the tool is to find out "whether a patient's health has contacted the needs of an expert doctor, if so, how urgent it is, and why it is so urgent."

DeepMind claims that its products can not only provide diagnosis, but also accurately explain its conclusions and the accuracy of the results, which is essential for health care professionals.

"for ophthalmologists, the effect of this product is surprising. You can see that AI has segmented each point, and there are about 6500 data points in the ultra-high resolution images created in this scan, "said Pearse Keane, a consultant ophthalmologist at Moorfields." We have to provide the same degree of rigour about how we verify the algorithms we use in medical devices, but I personally think ophthalmology will be the first medical profession to fundamentally change artificial intelligence. "

Two major medical strategies of DeepMind

In 2016, DeepMind set up the health department as part of its application team under the leadership of co-founder Mustafa Suleyman.

According to the Financial Times, the team's goal has been to develop and deploy artificial intelligence technology in the real world, including family doctor clinics and hospitals, but so far there are no commercial products and the department has not generated any revenue.

In November of 18, DeepMind announced that its health unit, DeepMind Health, and the team responsible for promoting "Streams" (the mobile APP that helps doctors identify and diagnose patients more quickly) would be adjusted and merged into Google's newly established "Google Health" division. Atomic company DeepMind Health will no longer exist as an independent brand.

Streams application

At the time, DeepMind's founder said it would be an "important milestone" for the company to transform its Streams application into an "artificial intelligence assistant nurse and assistant". The doctor "combines the best algorithm with intuitive design." The Streams app is currently being piloted in the UK to help healthcare workers manage patients.

However, DeepMind is not profitable in this respect, and an expert assessment panel developed by Alphabet questioned DeepMind and asked the latter to explain to the public what business model to take to make a profit.

This prototype of DeepMind's device may be able to answer this question. In the future, if all goes well, this product may become a breakthrough for DeepMind to make a profit.

The Medical Road of DeepMind

In addition to this eye disease scanning device, DeepMind has also made outstanding contributions to head and neck cancer treatment, AI diagnosis of breast cancer, and prediction of kidney damage.

In August 2016, DeepMind partnered with researchers from University College London School of Medicine to carry out a treatment program for head and neck cancer.

DeepMind said that because of the complexity of the human head, doctors need to spend at least four hours preparing a treatment for head and neck cancer, and they hope to automate part of the process through machine learning, reducing the preparation time to an hour.

In November 2017, DeepMind announced a new collaborative research project to try to use AI to diagnose breast cancer. In this research project, DeepMind will work with a number of medical research institutions led by Cancer Research UK. The main goal of this project is to apply machine learning to mammography, a routine examination of breast cancer, to improve its sensitivity and accuracy.

In February 2018, DeepMind established a medical research partnership with the U.S. Department of Veterans Affairs (VA).) DeepMind will analyze approximately 700000 anonymous medical records to find patterns that may be used to identify cases of acute kidney injury and common disease indicators to determine whether machine learning can accurately identify risk factors for deterioration and correctly predict the incidence of disease.

Related reports:

Https://www.ft.com/content/3de44984-9ef0-11e8-85da-eeb7a9ce36e4

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.