The future Dr House, TV’s complex diagnosis champion, will without a shadow of doubt be an artificial intelligence system. Software programmes have already attained the diagnostic levels of the top specialists and their accuracy is constantly improving. Does this mean however that we will have to hand over our entire digital ‘biography’?

Medical diagnostics: will patients have to sacrifice data privacy in return for better care?

Big Data and Artificial Intelligence (AI) technologies are making great strides in the medical field. Almost every week a new machine learning app proves its efficacy in diagnosing serious illnesses. IBM has published a great deal about the work the company has carried out with the Memorial Sloan Kettering Cancer Center in New York, where staff working in oncology ‘trained’ Watson up in this field. By compiling all patient data, IBM Watson helps oncologists to reach the right diagnosis and identify the best treatment for the patient. Since the pilot project in New York, ‘Watson for Oncology’ has been rolled out in Thailand and at six hospitals in India, and IBM recently announced that Watson is to be used in twenty-one hospitals in China as well. Computer technology is now of real assistance in compiling patient data, and is simply unbeatable when it comes to medical imagery.

‘Watson for Oncology’: IBM is introducing AI into an increasing number of hospitals worldwide. The tool can certainly save lives, but what about the threat to data privacy?

Recently, University of California, Los Angeles (UCLA) researchers demonstrated that combining two components – a deep learning computer programme and a photonic time stretch microscope –  enables its software to identify cancerous cells from a biopsy more efficiently than traditional methods, and at 36 million images per second! In 2014, researchers at the Canadian Institute for Advanced Research (CIFAR) showed that Deep Learning could indeed detect mutations in a patient’s genes that could lead to autism, colon cancer and spinal muscular atrophy.

AI in medicine goes back to the 70s

Using Artificial Intelligence techniques for medical diagnostic purposes is actually nothing new. The first experiments go back to 1972, with the AAPHelp software programme, but recent progress in AI, especially in image analysis, is now really pushing the boundaries of these systems. A couple of weeks ago, Google announced that its artificial intelligence subsidiary DeepMind was partnering with Moorfields Eye Hospital in London to diagnose and determine the correct treatment for common eye conditions such as age-related macular degeneration and diabetic retinopathy. Deep Learning algorithms developed by DeepMind will be ‘taught’ to read the 3D images of the retinas of patients taken using a medical imaging technique known as optical coherence tomography (OCT). A sample of a million anonymised images will enable the algorithm to learn to ‘read’ retinas and deliver a diagnosis. French startup DreamUp Vision has already demonstrated that this approach is very useful for diagnosing diabetic retinopathy.

"Our Deep Learning algorithm has learned how to trace this condition and how to carry out a diagnosis at various different stages," revealed Ekaterina Besse, CEO of DreamUp Vision at a recent Big Data event. "The algorithm rapidly detects complex shapes on the image of the retina and has understood how to build a diagnosis at a level of accuracy comparable with that of professional ophthalmologists," she told the audience.

DreamUp Vision’s algorithm has already attained a 97.5% diagnosis accuracy rate, compared with 96% to 98% for a trained ophthalmologist. The software is thus as efficient as a specialist but can be used by any healthcare professional – a general practitioner, the company doctor, a diabetics specialist or a nurse. "For the patient, it means detecting the condition early on and the diagnosis is immediate. There are an estimated 285 million visually impaired people worldwide and 80% of these cases could have been avoided if there had been an earlier diagnosis. Diabetic retinopathy is just the first step; in the future our software will be able to detect other illnesses that lead to visual impairment," predicts Ekaterina Besse. It is not hard to imagine just how useful such automated diagnostics systems could be in emerging countries where specialist doctors are rare and usually concentrated in major cities, or even in countries such as France, where the number of practicing specialists in certain fields is starting to diminish.

In the UK, Google DeepMind is re-inventing the practitioners’ job by using mobile devices with instantaneous diagnostic and collaboration apps linked to a given patient’s file – a very different vision from the French Personal Medical File system

Artificial Intelligence is definitely now banging on hospital doors and Big Data will play a major role in the future of medical care with the concept of P4 medicine – i.e. medicine that is predictive, personalised, preventive and participatory. An analysis of all the biomedical data on a patient, plus also data from a range of sensors including activity trackers will enable the risks of certain illnesses to be anticipated via predictive algorithms, and also help to create treatment tailored to each patient in order to increase the effectiveness of the chosen medicine and minimise any side-effects.

AI still scary, though…

So, given this surge of innovation flowing from digital technologies, the Internet and tech startups, have the rules on data privacy in the health field become obsolete? In an article published by not-for-profit media outlet The Conversation, Owen Johnson, a Senior Fellow at the University of Leeds in the UK who is conducting research into medical information systems, points up issues that could put the brakes on the use of these potentially life-saving systems. He writes about how slow the medical profession has been to adopt AI, pointing out that forty years after it was invented, “AAPHelp is still not in routine use.” He senses that people tend to be afraid of AI, while patients do not like the idea of being reduced to a set of statistics. Earlier this year a scandal erupted in the UK when it was revealed that the UK National Health System had given Google DeepMind access to 1.6 million patients’ medical records, including names and medical histories. The initial objective was to feed the algorithms of its kidney disease diagnostic mobile app, but an error enabled DeepMind to access the complete files.

Owen Johnson argues that “the hard slog is not creating the algorithms, but the patience and determination required to conduct careful work within the restrictions of applying the highest standards of data protection and scientific rigour.” He goes on to say: “At the University of Leeds Institute for Data Analytics we recently used IBM Watson Content Analytics software to analyse 50 million pathology and radiology reports from the UK. Recognising the sensitivities, we brought IBM Watson to the data rather than passing the data to IBM.”

France determined to create a proper structure

Meanwhile US high-tech giants have become very interested in health data. In acquiring Truven Health Analytics for $2.6 billion, IBM got its hands on 215 million patient profiles, an excellent playing field for its Watson AI system. However, from a French viewpoint the very idea of a private corporation gaining control over patient data is pure heresy. France’s Office of Shared Health Information Systems, ASIP Santé, is keeping a watchful eye on the situation. This agency, which is part of the Ministry of Social and Health Affairs, has published common guidelines for eHealth initiatives, containing a list of requirements enabling a better, more precise picture to be obtained of the players proposing to provide solutions in this field. "These guidelines provide legal and technical indicators, enabling everyone to understand both the current framework and the way it’ll evolve in the future", explains Pascale Sauvage, Policy Director at ASIP Santé. “Big Data opens an amazing range of possibilities for patient care, but we want to go forward step by step. One major step we have taken is to set up a framework to ensure the basic data used is reliable, quality data. Big Data is a fine thing but if we race ahead with applications there’ll be no guarantee whatsoever of data quality and valuable use cases will simply not appear,” he warns. Before being allowed to host health data on French patients, a provider must apply to the letter the common guidelines drawn up by ASIP Santé, and the agency will examine the application file in conjunction with CNIL, another government body whose task is to enforce the law on data privacy in relation to computer and digital files, before the Health Minister gives her final approval. 

Deep Learning algorithms are already the equal of trained specialists in interpreting medical images

In addition to drawing up rules for securing data storage and transfer, ASIP Santé is working on creating a precise structure for medical data, a legacy of the Personal Medical File system set up in France as long ago as 2004. Faced with the advent of Big Data, Pascale Sauvage is a strong proponent of a structured approach to the information. "If people are going to work on data on a large scale, the data will need to be as homogeneous as possible and if you’re going to compare one set of data with another they will need to be as structured as possible,” he argues, explaining: “To ensure interoperability, data must be structured by specific pathology so that all software in use can share common datasets. It’s going to be much harder if the data is not properly structured as automated language processing is the long-term future. We’re thinking about what needs to be put in place to facilitate automated language processing going forward but that will take a few decades.”

Will data protection hinder innovative solutions?

Notwithstanding this very French approach, with a hyper-secure, centralised and structured system, tech startups from all over the world and major players such as IBM and Google are creating new innovative solutions almost on a daily basis. Could the French approach hinder innovation? Is there a risk that under the pretext of protecting French patients, France’s authorities might end up depriving them of the opportunity to benefit from these innovations?  “The guidelines are not an obstacle in themselves,” argues Pascale Sauvage, pointing out: “Companies will need to incorporate them in advance, as soon as they come up with a new service. Today in France there are a number of hosting specialists that are entitled to host health data on behalf of a startup, which would solve the certification issue as easily as that. The main difficulty will be to incorporate the specific requirements of the guidelines right at the start of the project." It remains to be seen whether the Silicon Valley startups and US web giants will be persuaded to bow to the requirements of French medical data regulations.


By Alain Clapaud
Independent journalist specialising in the new technologies