You
might have heard of DeepMind last year, when they invented a program
that could beat human players in the game Go. The British artificial
intelligence company, now owned by Google, has been pushing the
boundaries of algorithmic learning research for some time now. I was
interested to learn that currently their main field of application is
healthcare. An entire division of the company is working on a range of
challenges with medical data, from interpreting medical images to
integrating patient data, all under the banner of DeepMind Health [1].
This work has been taking place in collaboration with a number of UK NHS foundation trusts. For example, a project aimed at detecting and predicting ocular degeneration from digital eye scans is the outcome of a collaboration with Moorfields Eye Hospital London, who have granted access to their database of anonymized digital images of the eye. Similarly, patient CT and MRI scans from University College London Hospital are being used in a machine learning approach to improve treatment planning for head and neck cancers.
How Big Data Helps Medical Professionals
Their largest project has been a collaboration with the Royal Free London NHS foundation trust and Imperial College Healthcare NHS trust in developing a mobile app to provide real-time patient information to nurses and clinicians. Called Streams, this app intends to speed up communication and decision making in hospital environments by replacing a number of older solutions relying on papers, fax, or pagers. The intention is to consolidate a patient’s medical results within a single interface where data-driven alerts can be sent out as soon as there is any indication of a problem, and actions can be decided upon by relevant health workers. The current focus is on a specific disorder, acute kidney injury, where such an approach is presumed to be particularly promising, but obviously the vision extends far more broadly.
From this technology, it is not difficult to imagine a future where data from multiple continuous bio-monitoring sources could be integrated so that patients, or any individuals, could be diagnosed and monitored in real-time. Glimpses of this path can already be seen with existing technology like continuous glucose monitoring devices, which provide continuous real-time blood-glucose measurements to diabetics, the data from which can then be accessed (and sometimes shared) via mobile apps [2]. Perhaps even more seemingly mundane biological data could be insightful from a medical perspective - think what information a Fitbit might reveal if state-of-the-art machine learning were applied to its data. Integrating all of these varied sources of information together to generate a comprehensive and detailed medical picture of an individual is surely something DeepMind Health have thought about as well.
This whole idea rests upon the ability for a private company to access potentially sensitive patient medical data, and DeepMind were doing just this, often without patient consent. Predictably, this drew some criticism. DeepMind responded earlier this year in an interesting way. Rather than attempting to seek patient consent, they instead took an approach of transparency by announcing their development of a data-logging process - the verifiable data audit.
Transparency vs. Data Privacy
The idea is that a record of all interactions with patient data will be generated and saved, with a log of who was accessing an element of data, when, and for what reason. This record will be automatically updated and stored in a semi-decentralized manner that has been likened to blockchain, with records in a distributed network of healthcare institutions such as hospitals. The data is structured so that any time it is accessed or changed this will be immediately recorded. As such, guidelines can be put in place to ensure that the data isn’t used in unauthorized ways. It also means that all access of data will be traceable forever in a way that should be tamper-proof.
At the epicenter of all this technological innovation is data - as the 21st century is starting to teach us, data is powerful and data is valuable. So where there is a question of data, there is always a question of privacy trailing close behind. If we can imagine these technologies, we must also be able to imagine a future where our most intimate biological details are shared widely and accessible to many. The price of a detailed understanding of our own body is perhaps our privacy in that matter.
In this discussion, DeepMind are not strictly advocating privacy - rather, they are advocating transparency. Yes, your data will be accessed by many people, but with a strict record of who and for what purpose, which will ideally necessitate adherence to data sharing guidelines. Before patients themselves have access to their own records, this will likely still sit uncomfortably for many people. Until then, the success or failure of initiatives like DeepMind Health will determine how willing we are as a society to invest our data in our health, and to whom.
[1] http://bit.ly/2nfaDQ2
[2] http://bit.ly/1QM9PKF
by James Kerr, PhD Student AG Sterzer
this article originally appeared June 2017 in CNS Volume 10, Issue 2, Digital Health and Big Data
This work has been taking place in collaboration with a number of UK NHS foundation trusts. For example, a project aimed at detecting and predicting ocular degeneration from digital eye scans is the outcome of a collaboration with Moorfields Eye Hospital London, who have granted access to their database of anonymized digital images of the eye. Similarly, patient CT and MRI scans from University College London Hospital are being used in a machine learning approach to improve treatment planning for head and neck cancers.
How Big Data Helps Medical Professionals
Their largest project has been a collaboration with the Royal Free London NHS foundation trust and Imperial College Healthcare NHS trust in developing a mobile app to provide real-time patient information to nurses and clinicians. Called Streams, this app intends to speed up communication and decision making in hospital environments by replacing a number of older solutions relying on papers, fax, or pagers. The intention is to consolidate a patient’s medical results within a single interface where data-driven alerts can be sent out as soon as there is any indication of a problem, and actions can be decided upon by relevant health workers. The current focus is on a specific disorder, acute kidney injury, where such an approach is presumed to be particularly promising, but obviously the vision extends far more broadly.
 |
| Image via pixabay |
From this technology, it is not difficult to imagine a future where data from multiple continuous bio-monitoring sources could be integrated so that patients, or any individuals, could be diagnosed and monitored in real-time. Glimpses of this path can already be seen with existing technology like continuous glucose monitoring devices, which provide continuous real-time blood-glucose measurements to diabetics, the data from which can then be accessed (and sometimes shared) via mobile apps [2]. Perhaps even more seemingly mundane biological data could be insightful from a medical perspective - think what information a Fitbit might reveal if state-of-the-art machine learning were applied to its data. Integrating all of these varied sources of information together to generate a comprehensive and detailed medical picture of an individual is surely something DeepMind Health have thought about as well.
DEEPMIND HAVE TAKEN A TRANSPARENCY APPROACH
This whole idea rests upon the ability for a private company to access potentially sensitive patient medical data, and DeepMind were doing just this, often without patient consent. Predictably, this drew some criticism. DeepMind responded earlier this year in an interesting way. Rather than attempting to seek patient consent, they instead took an approach of transparency by announcing their development of a data-logging process - the verifiable data audit.
Transparency vs. Data Privacy
The idea is that a record of all interactions with patient data will be generated and saved, with a log of who was accessing an element of data, when, and for what reason. This record will be automatically updated and stored in a semi-decentralized manner that has been likened to blockchain, with records in a distributed network of healthcare institutions such as hospitals. The data is structured so that any time it is accessed or changed this will be immediately recorded. As such, guidelines can be put in place to ensure that the data isn’t used in unauthorized ways. It also means that all access of data will be traceable forever in a way that should be tamper-proof.
At the epicenter of all this technological innovation is data - as the 21st century is starting to teach us, data is powerful and data is valuable. So where there is a question of data, there is always a question of privacy trailing close behind. If we can imagine these technologies, we must also be able to imagine a future where our most intimate biological details are shared widely and accessible to many. The price of a detailed understanding of our own body is perhaps our privacy in that matter.
In this discussion, DeepMind are not strictly advocating privacy - rather, they are advocating transparency. Yes, your data will be accessed by many people, but with a strict record of who and for what purpose, which will ideally necessitate adherence to data sharing guidelines. Before patients themselves have access to their own records, this will likely still sit uncomfortably for many people. Until then, the success or failure of initiatives like DeepMind Health will determine how willing we are as a society to invest our data in our health, and to whom.
[1] http://bit.ly/2nfaDQ2
[2] http://bit.ly/1QM9PKF
by James Kerr, PhD Student AG Sterzer
this article originally appeared June 2017 in CNS Volume 10, Issue 2, Digital Health and Big Data
