[ad_1]
SEBASTIAN KAULITZKI / SCIENTIFIC PHOTOGRAPHIC LIBRARY / WIRED
In 2017, DeepMind began testing a new application with the Royal Free Hospital in London. This program, called Streams, was intended to help clinicians identify and monitor acute kidney injury – a condition linked to 100,000 deaths in the UK each year. But unlike most of the works that make the front page of DeepMind, Streams does not contain any trace of artificial intelligence.
Instead, the app collects medical information, such as blood test results and vital signs, and informs clinicians when a patient's kidney health is deteriorating, using a well-established formula. to evaluate kidney function. DeepMind has now provided early indications that the use of artificial intelligence could be a much better way to badess if a person is at risk for AKI.
In an article published in the scientific journal NatureDeepMind researchers have shown that they have created a machine learning algorithm that can predict AKI for up to 48 hours. Using a large database from the US Department of Veterans Affairs, the DeepMind team built an algorithm to predict whether a patient would have AKI. In 90% of the most severe cases, the predictions of this algorithm were accurate.
Dominic King, Clinical Lead for DeepMind, hopes that using AI to predict patient deterioration would allow clinicians to intervene sooner. In the case of ARIs, rehydration, antibiotics or drug modification can help to restore patients' kidney function quite easily. "Currently, we recover these objects too late and patients suffer. We believe that these AI systems have a real opportunity to predict and prevent rather than just what is happening right now, namely that clinicians are almost fighting fires and solving problems that have already developed, "says King.
The artificial intelligence system was formed on more than 620,000 separate data points, which identified 3,600 of them that were good predictors of IRA. "The power of in-depth learning lies in the fact that it allows you to extract a lot of these signals automatically if you have enough data to provide it," says Nenad Tomašev, an engineer. senior researcher at DeepMind and co-author of the book. Nature paper.
However, deploying this type of AI system will first require training and testing on much more diverse datasets. This latest study was conducted using historical data, taken between 2011 and 2015, and therefore was not used to monitor patients in real time. And although the total dataset contains more than 703,000 patients, only 6.32% were women, which means that the AI system was less effective at predicting the AKI when 39, it was tested on patients.
"Although this algorithm would work well for the [Veterans Affairs] population, we would not propose that without additional training and validation that you would use it elsewhere, "says King. "There is a lot of talk and work still to be done to deploy this in a real-life context," he says, believing that DeepMind is still in at least a year before starting to test the algorithm. In real clinical environments. .
Why does Google consume DeepMind Health? This scares the privacy experts
And that's where the streams come in. In the end, King claims that Streams' interest is in proposing such AI-based predictions – not just for the AKI, but for others as well. conditions such as sepsis, acute liver failure and diabetes. complications.
That's why DeepMind wrote three other articles – published the same day as the Nature paper – evaluate the usefulness of Streams at the Royal Free Hospital in London. An agreement between DeepMind and the Royal Free NHS Foundation Trust was concluded in 2017 to violate the data protection law. The agreement between the two has since been revised.
The studies paint a mixed picture of the success of Streams so far. The application could not be related to an improvement in renal function or to a number of other parameters used to determine renal health. But this, says King, is not totally surprising. This version of Streams always uses the NHS-approved algorithm that estimates the risk of AKI depending on the level of waste called creatinine in the blood. The problem is that creatinine levels can reach many hours after the installation of AKI. The ultimate goal of DeepMind is to replace or supplement this algorithm with a variation of the system presented in the Veterans Affairs study.
And while the application has not improved kidney health, it seems to make it easier and faster to treat patients at risk of kidney damage. Kidney specialists using the application have examined urgent cases in less than 15 minutes, compared with 4 hours for those who did not use the application, which means that they do not need to be used. have forgotten that 3% of cases of AKI.
DeepMind also estimates that Streams reduces the cost of admitting a patient with an AKI of £ 2,123 on average, although this figure does not take into account the cost of supplying Streams or the cost of dialysis. long-term in untreated patients. . King says he is convinced that if ever this was implemented in the NHS, the Streams would eventually be more profitable than the current AKI treatment approach.
This will determine whether the NHS and other health care systems wish to expand their use of Streams. By involving its healthcare providers with its technology, DeepMind attempts to capitalize on three key arguments: its technology improves patient outcomes, is useful for clinicians and patients, and does not increase the costs of patients. health care per patient.
The feeds already seem to tick boxes on the last two points and although his progress on the first one is yet to come, the Veterans Affairs experience suggests that a version of Streams based on the latest two points suggests that IA could have a significant impact on the health of patients. And while DeepMind's interest so far has been decidedly unbusiness, it seems that the company is slowing down more and more to start selling its products to healthcare providers.
This could be related to the announcement made in November 2018 that DeepMind Health should be part of Google, under the supervision of Google Health Vice President, David Feinberg, who took office in January 2019. Up to date now, Alphabet, the parent company of DeepMind, has kept The London company is distant, but the integration of DeepMind Health in Google suggests that the team will start to put more emphasis on the marketing of its creations.
Working more closely with Google could also bring additional benefits to Streams, King says. If Google knows what to do, it controls large amounts of information in an easily searchable manner – which will be essential for Streams if this information is eventually extended to other conditions. "The combination of these elements allows us to realize a much more complete and exciting project," he said.
More beautiful stories from WIRED
?️ How to exploit Google Photos for your messy photos
? Heat waves make people more violent, angry and grumpy
? England has an ambitious plan to eliminate smoking by 2030
It's time to drop Chrome for a priority web browser for privacy.
? A vaccine for Alzheimer's is about to become reality
Get the best tech deals and the latest gadget news in your inbox
[ad_2]
Source link
