AMRC Digital Health Bulletin

25 April 2019

Following last month’s announcement, NHSX has started taking shape, with Matthew Gould announced as its CEO and plans such as digital experts collaborating with mental health and cancer policy teams being set out. (3 April, 4 April) It’s not just the NHS that has seen exciting developments this past month though…

New in health tech

• An MRI-software system has allowed detailed 3D images of babies’ hearts to be created while they are still in the womb. This means that congenital heart diseases can be detected early-on and contingency plans for treatments, operations and parental support can be organised prenatally. A series of 2D MRI images from varying angles (blurry due to the high frequency BPM of the tiny heart and of little use in isolation) are combined by the sophisticated software that corrects for the heart movement to create the 3D image. (22 March)
• An early-stage startup is investigating whether a custom-made tampon could be worn for 2 hours and then sent to a lab in order to pick up early markers of endometriosis and in the future potentially other disorders such as cervical cancer. (1 April)
• This month news broke that Apple had filed patents potentially indicative of plans to develop more advanced health sensors. The first outlined solutions to measure light in order to alert the user to air pollution, while the second was a system to detect chemicals in the air, with one suggested application of this being that it could ‘smell’ a user’s sweat to detect if their blood sugar level is low. (4 April)
• A new potential treatment has been identified for treating the walking difficulties experienced by some Parkinson’s patients. It involved fitting an electrical implant to the spine in order to stimulate the feedback loop which carries signals from the legs back to the brain to communicate that a movement has been executed. The treatment shows promise in the initial small-scale study, enabling patients to walk that had previously experienced regular falls or needed a wheelchair. (22 April)

Collaboration to drive innovation

• The Consumer Technology Association has launched an AI in healthcare initiative that includes more than 30 organisations (including major tech companies like Google, Fitbit and IBM). The group will explore issues such as trustworthiness, ethics and bias to develop standards and best practice recommendations in hopes of advancing the field. It will be interesting to see how this develops. (4 April)

Leveraging data & AI

• Robots could offer a solution for patients with long-term conditions in need of consistent medical attention, offering more full-time (and perhaps even better) care. This article discusses the case of Rayfield Byrd and his experience with a robot that helps support him living alone following congestive heart failure and having Type 2 diabetes. The robot monitors key symptoms (e.g. shortness of breath, mood and anxiety) and behaviours (medication adherence and exercise) by ‘speaking’ with him. Byrd hasn’t missed any doses of his medication, has quit smoking, lost weight through regular exercise and avoided having his leg amputated due to plaque in a vein. (21 March)
• A toolkit of 13 pre-trained models to assist with medical imaging tasks have been released by Nvidia to do assist radiologists with common tasks. The toolkit also offers tools intended to aid in building and training your own models and sharing them. For example, ‘transfer learning’ allows deep learning algorithms to be customised e.g. to local demographics data and imaging devices. This allows doctors to create their own models for their patients without having to move or share patient data, and is said to require 10 times less data than starting an algorithm from scratch. Along a similar vein, researchers from New York University have developed a machine learning algorithm for identifying malignancies in mammograms that they have made open source. They claim that their pretrained weights will allow others with access only to smaller datasets to train models successfully. (19 March, 21 March)
• The security and privacy of healthcare data, in an increasingly data-driven healthcare space, is a key priority for many patients, with more and more organisations focussing in on this issue. Blockchain is being used to empower people to decide when and how their healthcare data is used, and CoverUS is taking this a step further by taking the profits that would otherwise be absorbed by third-party data brokers and passing these directly onto patients through their blockchain app. Another way to tackle privacy of data is by running AI locally. For example, this AI ‘radiology assistant’ assesses the likelihood of 14 different diseases based on a chest X-ray, and does so using the doctor’s browser. This means it is cheap to run as expensive, centralised servers are not needed and no data is transferred off the device. (22 March, 1 April)
• Following on from a Tuberous Sclerosis Complex (TSC) Priority-Setting Partnership that identified TSC-associated neuropsychiatric disorders (TAND) as the top research priority, the Tuberous Sclerosis Association is co-funding a data collection project in this area. An app will be developed that will allow people to self-report how they are affected by TAND to enable research to improve treatments. (4 April)
• AI has applications not only in clinical care but also within medical research and can alleviate labour-intensive processes to free up experts’ time. For example, researchers have created an open-source deep learning algorithm that is able to identify individual neurons within calcium imaging recordings of neuronal firing as accurately as neuroscientists but in a fraction of the time: 20-30 minutes processing for a 30-minute video compared to 4-24 hours for a researcher. (12 April)
• It is argued that the use of AI to complete voice-based diagnostics could mean that diagnostics would be done seamlessly as people go about their day-to-day life, with less clinician or patient effort. For example, a paper this month detailed an AI tool that was able to delineate those with and without PTSD with 89% accuracy by identifying key speech patterns associated with it (e.g. less clear speech and a metallic tone). (22 April)

Ideas from abroad

• In Israel, the four Health Maintenance Organisations that provide all the country’s healthcare have been using the same electronic health records platform for the past two decades, enabling a huge wealth of data to be amassed around patients, conditions and treatments. The positive results of this joined up system are already being seen. For example, a big data analysis has meant routine blood tests can highlight risk of colorectal cancer in its earliest stages based on variations that would otherwise be viewed as 'within the norm’. Another use case is in determining the ideal dosage of antidepressant for a patient almost instantly, which is done by matching patients on genetic information and data such as symptoms, conditions and lifestyle. This digital health ecosystem has also brought about significant attention and investment. (26 March, 16 April)
• The FDA has published a discussion paper (and request for feedback) that sets out a framework for how they could potentially address AI algorithms and how to deal with the fact that algorithms are often iterated (while existing guidance requires resubmission for clearance if major changes are made to software). The proposed approach is that submissions would include plans for anticipated modifications via ‘SaMD Pre-Specifications’ and ‘algorithm change protocol’. The former outlines the changes the company anticipates making, while the latter sets out details of the processes that will be followed to implement these changes. (2 April)
• The World Health Organisation have released a draft Global Strategy on Digital Health, which includes three guiding principles, four strategic objectives and a framework for action. The strategic objectives aim to identify areas WHO, its partner and stakeholders should focus on, and they are requesting feedback within the public consultation (closes 30 April) from stakeholders in the digital health arena from across all sectors. (8 April)
• This report outlines a project that investigated the use of digital technologies to improve mental health services in the US and Australia. It summarises a number of case studies using technology within service design and delivery, prevention/self-help/peer support, digital phenotyping, supporting in-house innovation and enabling research and based on learnings from these provides five key recommendations for the NHS. Do you think a global perspective would be useful for other aspects of health to highlight areas of improvement for the NHS? (14 March)

Government, NHS & policy

• The Centre for Data Ethics and Innovation has released its first work programme and strategy. Alongside outlining how the group will work and engage with stakeholders and its governance structure, a number of key outputs are outlined. These include an analysis of opportunities and risks, a 2020 State of the Nation report focussing on the use and governance of data-driven technology in the UK, and two in-depth reviews focussing on online targeting and algorithmic bias. (21 March)
• The projects receiving a share of the £9 million Digital Health Technology Catalyst were announced, including a sleep apnoea device, low-cost brain imaging technology to assay dementia risk and conversational AI to assess patients’ health status following cataract surgery. (28 March)
• The EU High-Level Expert Group on AI published its ‘Ethics guidelines for trustworthy AI’, highlighting 3 key components for Trustworthy AI: lawful, ethical and robust. (9 April)