DeepMind Healthcare Unit To Merge with Google To Advance AI Healthcare Research

Categories

As recently announced, Google is bringing the DeepMind Healthcare Unit into its fold to bolster its efforts in the competitive field of AI healthcare research. DeepMind’s health subsidiary will now be part of Google Health spearheaded by David Feinberg, the former CEO of Geisinger; and are said to be part of a broader effort by Google to boost collaboration and communication among health projects at Alphabet, which are currently scattered across the organization. The DeepMind and Google healthcare teams will be combined to help them become the “AI-powered assistant for nurses and doctors everywhere”.

DeepMind, one of the world’s foremost AI initiatives headquartered in London with additional research centres in Edmonton and Montreal,  was acquired by Google in 2014 and is part of the Alphabet group. However, the AI company, including its health division has been operating as an independent brand until now when it will just be part of Google.

AI can generally be divided into “task AI” and “general AI” (i.e. more flexible intelligence that more closely resembles our own). The healthcare industry has witnessed a growth in task-oriented decision-making via AI, especially in terms of administrative procedures. However, clinical decisions led by a more general AI have so far largely remained out of reach. DeepMind, however, falls more into the general AI category.

DeepMind announced the news of the merger in a blog post that focuses on its mobile app, Streams, which helps doctors and nurses “deliver faster, better care to patients”. DeepMind says that one of the reasons “for joining forces with Google in 2014 was the opportunity to use Google’s scale and experience in building billion-user products to bring our breakthroughs more rapidly to the wider world”.

DeepMind’s health division has made major strides in healthcare AI research since its founding in 2010 in multiple different areas from detecting cancer radiotherapy to treating eye disease. There were two goals in place from the start: (i) “to make a practical difference to patients, nurses and doctors and support the NHS and other healthcare systems”; (ii) to make DeepMind Health a self-sustaining initiative, through hospitals choosing to pay us for our software if they think they can have a positive impact on clinical outcomes and experience”.

DeepMind’s Streams

Streams is a DeepMind mobile app used in the UK’s NHS hospitals in order to address what clinicians term “failure to rescue” i.e. when a patient is not treated in an adequate amount of time. Many thousands of people in UK hospitals die from preventable conditions such as kidney injury or sepsis as the warning signs are often not detected and acted on in time.  Streams was built to address this problem.

The app holds the important medical information of patients in one place, such as their blood test results so that clinicians at partner hospitals can see serious issues when patients move around. If an issue is detected, Streams will send an urgent smartphone alert to “the right clinician”, in addition to information about prior conditions so an immediate diagnosis can be made. Streams also enables clinicians to instantly review vital signs such as blood pressure and heart rate and record the observations directly into the app. It does so by integrating “different types of data and test results from a range of existing IT systems used by the hospital”.

Nurses at The Royal Free Hospital where Streams has been deployed said it was saving them up to two hours each day. It has mainly been used at The Royal Free to help clinicians better detect and treat acute kidney injury, a condition which has been tied to 40,000 deaths in the UK each year, a quarter of which NHS England estimates could have been prevented. It costs the NHS over £1 billion – more than the amount used annually to treat breast cancer.

DeepMind continues to update Streams with new features that enable secure communication between doctors for instance, and is working on developing the app into including alerts for more avoidable conditions such as organ failure and sepsis.

In June 2018, DeepMind disclosed that deals to use the app in another 10 hospitals inside the NHS had been made.

Privacy Controversy over Streams

Also in 2016, Streams found itself at the heart of a controversy around the use of private data when New Scientist reported on a data-sharing agreement between DeepMind and the NHS, which showed their collaboration had gone way beyond what had been publicly announced. The agreement showed that DeepMind had been granted access to healthcare data on 1.6 million patients who had passed through three London hospitals run by the Royal Free NHS Trust. The information includes data on people who are HIV positive, in addition to details of drug overdoses and abortions. Sam Smith, leader of health data privacy group MedConfidential, said the document showed that DeepMind was gaining access not only to records about kidney function directly related to the app’s function, but to a wider array of historical medical records. These records include logs of day-to-day hospital activity. Since the scandal broke in the UK, DeepMind has worked hard to reassure the public of its commitment to patient privacy. In the US, however, health systems routinely share data with outside vendors such as analytics firm under a framework permitted by HIPAA.

Other DeepMind Healthcare Initiatives

Breast Cancer Screening – Collaboration between UK and Japan

A research partnership between Imperial College of London and DeepMind in the UK to assist clinicians in more accurately diagnosing breast cancers on mammograms more effectively and quickly earlier this year saw The Jikei University Hospital in Japan join in these efforts. The Jikei University involves analyzing historic, de-identified mammograms from circa 30,000 women performed at the hospital between 2007 and 2018. AI technology is being used to investigate if the technology is able to “detect signs of cancerous tissue on these X-rays more effectively than current screening techniques allow”. The goal is to use data from multiple different groups as by under-representing certain groups, whether age, ethnicity or gender, can lead to the creation of technology which doesn’t best meet all groups’ interests. In the instance of breast cancer, there are often considerable differences in breast density between different ethnic groups. Bias in the AI system and training process can lead to breast cancers being misdiagnosed or even missed entirely.

Diagnosis of Sight-Threatening Eye Disease – UK Research

DeepMind worked in collaboration with Moorfields Eye Hospital in London to explore whether its AI system was able to quickly interpret eye scans from routine clinical practice. The research partnership determined that the AI system was able to “correctly recommend how patients should be referred for treatment for over 50 sight-threatening eye diseases as accurately as world-leading expert doctors”.

The DeepMind system takes a “novel approach” to the problem of “the black box”, meaning it is hard to understand why an AI system makes a recommendation. In healthcare, this is problematic for both patient and clinician. The DeepMind system combines two neural networks that address this issue, providing visualizations and recommendations as percentages in ways that reassure the clinician in the system’s capability for analysis.

Takeaways

The next question is how will Google apply AI in healthcare, and if in so doing, the company will hold its own with the other big tech giants (Microsoft, Amazon, Apple) all also making a foray into AI and healthcare. DeepMind Health being folded into Google is certainly worth closely watching. Will the collaboration enable commercialization on a large scale? As HealthDataManagement wrote, If they can do this commercialization successfully, and translate DeepMind’s advances into scalable decision support for patients and providers, there is tremendous potential.” If new technology can help clinicians make more accurate analyses, and get quicker treatment for those patients who most need it, we will see it embraced the world over more and more.

Scroll to Top