AI diagnostics: innovation race is on as research confirms accuracy

null

Medical diagnosis has been a traditional area of application for artificial intelligence (AI) systems for decades. However, recent technological advances, along with promising outcomes from clinical trials, have given fresh impetus to this dynamic area of research and development.

Once regarded as a test bed for AI research and development, a series of well-publicised clinical trials linked to medical diagnosis has confirmed that, in some areas at least, the technology is application-ready and capable of making a real contribution to improving the delivery of health services. Among the recent successes, AI systems have demonstrated their ability to diagnose and predict disease with a high level of accuracy, particularly in applications that involve the intelligent analysis of medical imaging, such as mammography or retinal imagery.

Recognising the potential of this technology, the Government has recently announced plans to invest £50 million in the creation of five new medical technology centres, which are due to open in 2019. These centres will be located at universities and NHS facilities across the UK and will support the development of AI systems for medical diagnosis, with the aim of improving the efficiency of health service delivery. The funding is being provided by the Industrial Strategy Challenge Fund and it is hoped the initiative will encourage further investment in the field.

A detailed analysis of patent-publication data in the US and Europe indicates that AI-related innovation activity linked to health or medical diagnosis grew by nearly 400 per cent between 2013 and 2018. Based on this analysis, the rate of increase is expected to continue over the next few years.

While most research scientists would probably have expected to see some acceleration in AI-related innovation, the growth in patent publications linked to medical diagnosis is significant. To put this into perspective, US and European patent publications involving the use of AI increased by 300 per cent in the same period, whilst those linked to medical diagnosis remained largely constant.  In addition, approximately 1 per cent of the total number of medical diagnosis related patents published in 2013 were related to uses of AI within the field.  By November 2018, this proportion had increased to over 7 per cent.

As good as the experts

With AI innovation advancing rapidly, the list of potential applications in the field of medical diagnosis is expanding. In one of its earliest applications, the technology has been used successfully in the field of ophthalmology to screen retinal fundus images in order to identify diabetic patients with eye conditions requiring medical intervention. Trials have shown that these intelligent systems produced reliable results and performed at least as well as individual ophthalmologists.

Building on this work, Google AI and Verily Life Sciences, another subsidiary of Google’s parent company Alphabet, are spearheading research to screen retinal photography in a similar way, whilst using algorithms to process other patient data, in order to assess whether they are at risk of a stroke or cardiovascular event, such as a heart attack. Google subsidiary, DeepMind also claims to have technology capable of diagnosing a range of eye conditions with a high level of accuracy within a matter of seconds.

Based on an article published originally at China Daily, BioMind, a Chinese company specialising in deep tech innovation for medical applications, has developed an AI system capable of diagnosing brain tumours and predicting the expansion of brain haematomas, with a high level of accuracy. The technology, which was developed by the AI Research Centre for Neurological Disorders at Beijing Tiantan Hospital and a research team at Capital Medical University, was able to make correct diagnoses in 87 per cent of 225 cases in about 15 minutes. In trials conducted at the hospital, the technology outperformed a team of 15 doctors from across China who diagnosed the same cases with 66 per cent accuracy in about 30 minutes.

Another fast-developing area of application is the use of AI systems to analyse images of skin lesions in order to identify cancer.  An AI system developed by Stanford University has been trained on a dataset of over 125,000 images of benign and malignant skin lesions.  The system, which utilises a deep convolutional neural network, has been demonstrated to achieve performance comparable to expert dermatologists.  The trained system is expected to be deployed as a smartphone app and so help to lower the cost and time-to-diagnosis for patients.

The importance of speed to market

As the Stanford system shows, AI systems have particular strengths in that they can learn from large datasets and identify patterns which can assist in medical diagnosis. This means that the technology is especially effective in areas such as pathology and radiology, which rely on an ability to process data quickly in order to spot patterns or irregularities. Despite their increasing reliability however, these systems are unlikely to replace the clinical pathologist or radiologist entirely. For the foreseeable future at least, AI systems will be most effective when deployed alongside clinicians, helping to streamline the clinical decision process and offering fast and effective second opinions.

Whilst some issues remain, the promise of AI in the field of medical diagnosis has inspired a mini health-tech boom, as a growing number of start-ups enter the sector. With the backing of venture capital investment, some of these companies have been quick to move their prototypes and early-stage innovations into real world trials. From an intellectual property perspective, it is especially important that these businesses have patent protection in place. Not only could this help them to secure the finance they need to get started, it will also prevent rivals from copying their ideas or beating them to market with a system that can be used in the same or a similar way. In the longer term, building a robust IP portfolio could also position the business favourably for potential mergers or acquisitions as the AI market begins to mature.

Backed by positive research outcomes and strong investor interest, finding new ways to use AI systems to speed up the diagnosis of disease is likely to remain a priority for health-tech innovators for many years to come. Whilst speed to market is important, the businesses joining this innovation race must ensure their inventions have the protection required to realise their commercial potential.

Karl Barnfather, partner and chairman, Withers & Rogers
Harry Strange, patent attorney
Image Credit: PHOTOCREO Michal Bednarek / Shutterstock