Robot surgeons who do not tremble, blind people who regain their sight, prolonged life expectancy, cancers detected before the first symptoms… new technologies perform miracles, unimaginable a few years ago. Between AI, machine learning, robotics and implants, medtech improves patient health and saves lives. Trainer, expert in artificial intelligence and robotics, Cédric Vasseur takes the pulse of medicine 3.0 and answers our questions.
Health is one of the areas where we hope to make the most progress thanks to artificial intelligence. The latter took off in the 2010s, in particular thanks to advances in deep learning which imitates, via machine learning techniques, the learning method used by human beings to acquire certain knowledge. It all started above our heads, in space.
Could the conquest of space be at the origin of the medical innovations of recent years?
Perfectly! Like other sectors, medicine follows scientific advances. The artificial heart, insulin pumps or MRI, magnetic resonance imaging, are commonplace today. And yet these innovations come from the conquest of space.
The ventricular pump used in artificial hearts is derived from the fuel pumps of the American space shuttle from the early 2000s. The first French artificial heart was also co-developed by engineers from EADS and Matra in 2008. As for MRI, if it was not invented by NASA, it was greatly improved by its engineers who needed to enlarge photos of the Moon!
In the same vein, the operation of insulin pumps is inspired by the robotic arm of the Viking probe launched in 1975. And the dialysis machines are an improvement on the fluid recycling systems of the Apollo missions. Robotics and artificial intelligence are only following this illustrious lineage.
What does AI bring to the healthcare sector?
Medical imaging is the first to benefit from advances in artificial intelligence. Artificial intelligence has been added to MRI machines and scanners to improve image quality and facilitate diagnostics. You should know that MRIs are extremely expensive devices, more than a million euros for the latest generation models. As a result, some companies are selling AI-based software solutions to update old MRI machines and improve their image quality, without the hospital needing to invest in a new machine.
In addition to improving image quality, AI will facilitate diagnosis. A radiologist may miss an anomaly, either because he cannot see it with the naked eye, or because he is tired after analysing thousands of images during the day. The machine, on the other hand, does not tire. According to a study published in 2018 in the prestigious journal Radiology, up to 25 % of cancers visible on a mammogram go undetected because of fatigue and lack of concentration. This is a worrying figure, especially as the amount of health data to be analysed is increasing exponentially: a radiologist sees up to 50,000 images per day, 40 times more than thirty years ago!
How does AI work? It uses machine learning : we feed the machine with thousands of radios in order to train it to detect anomalies autonomously. Thus, a PET scan is now capable of detecting the presence of metastases instantly and automatically. The technology is so effective that radiologists are gradually at risk of being replaced by these machines.
Is artificial intelligence helping doctors in other ways?
Yes, artificial intelligence has even advanced knowledge in radiology. In 2017, IBM created Watson, an artificial intelligence capable of analyzing different types of cancers. After viewing thousands of X-ray images, the AI made doctors understand that the border of an image's usual analysis area contained important information, such as the center of the tumor.
AI is used to improve access to knowledge. Doctors around the world share their information on which molecule to take depending on this or that symptom or disease. You have to imagine the amount of information to process: thousands and thousands of pages per day of studies and epidemiological reports. A job that a human being cannot assimilate, no matter how expert he or she may be. Recognized as an expert in oncology, IBM's Watson is fed day after day with articles from medical journals, biopsy results, laboratory tests, comparisons of molecules... When we ask him about a treatment, Watson will change its recommendation based on its latest updates and will recommend molecule Z rather than B.
Less impressive, but very useful to doctors, AI is used every day by doctors to type up medical reports. We used to have an army of secretaries typing up reports dictated by doctors. Today, Nuance, through its flagship product Dragon Naturally Speaking, enables doctors to use speech recognition based on deep learning and machine learning to transcribe their medical reports.
Can AI diagnose diseases before the first symptoms appear?
Yes, for example, IBM's Watson can predict the onset of breast cancer a year before it occurs. Other AIs are used for epidemiological predictions. We can thus follow the progression of the flu thanks to algorithms which scan social networks in search of people complaining of its symptoms.
Using the same type of technique, an AI from the Canadian company BlueBot detected the emergence of the pandemic on December 31, 2019 when it was then considered to be an epidemic of pneumonia confined to Wuhan, China. To do this, she used around a hundred data sets from the Internet (news sites, airline ticket sales, demographic, climatic and animal population data, etc.) to determine the threat, its geographic location and its propagation by providing a list of cities likely to be confronted with it in turn.
What do robots bring to medicine?
AI is not the only one making its contribution to medicine. Robot surgeons considerably improve the precision of surgical operations. The most famous is the American Da Vinci. This robot is equipped with numerous instruments (tweezers, scalpels, cameras, etc.) manipulated remotely by a surgeon present in the operating room.
AI improves robots: it makes it possible to automate certain actions and improve the working conditions of surgeons. When an operation goes badly and blood covers what the surgeon is doing, the machine can continue to make stitches on its own.
The precision of these robot surgeons is so great that their manufacturers show that they can put stitches on grapes without bursting them.
Other robots, still in development, are capable of automatically inserting a catheter. The procedure is very complicated and usually requires a nurse who knows how to detect veins.
And the implants?
When it comes to implants, current technologies are reminiscent of science fiction films or miracles: profoundly deaf people can hear, blind people can see.
Let's take the example of deaf people. They are surgically implanted with a cochlear implant, a set of small electrodes that send small electrical impulses into the cochlea. After the operation, they hear, certainly not like you and me. They have to go through a stage of rehabilitation and they will not be able to detect musicality easily. In the 1990s, implant recipients had to wear a device the size of a Walkman on their waist, with a cable that ran up under the flesh. Today, cochlear implants work by induction. An antenna is placed under the skin. A magnetic outer part contains the electronics and sits above the skin. You remove this mini-box and the person is deaf again. Now, we are able to place the implant directly on the nerve, or even implant electrodes directly in the brain. And it works !
For vision, the Second Sight company also uses implants. At the moment, only a few pixels appear in people's eyes. These few pixels allow them to gain autonomy since they can move around the street, see the signs, the sidewalks and the objects they want to grab.
Will AI and robots apply to everything?
No. Some areas of medicine are too complex for AI and robots. For example, in orthopaedic surgery, we tested a robot to operate on a foot. The machine made too many mistakes and was withdrawn from the market. What's more, certain procedures cannot be automated. All the medical and personal assistance professions that require psychology, imagination, communication skills and empathy are likely to remain in the hands of humans.
This does not mean, however, that these professions will not one day be assisted in whole or in part by robotics or other forms of new technology: robotic surgeons, robotic draughtsmen, 'intelligent' decision-making tools, etc.
What are the underlying problems?
These technologies are both dreams and fears. Fear that the machine will be the only one to make decisions, fear of its dehumanizing side. However, in medicine, many analyzes are already completely automated, such as PCR tests and blood samples. Humans nevertheless always have their place during sampling.
Robots can even improve the lot of certain people: dependent elderly people by making them more independent when it comes to their own toilet. These people feel less embarrassed by dealing with a machine.
The other limitation encountered by these technologies is at the data level. AI requires large quantities of data, and more importantly, high-quality data. This also raises the problem of data collection, consent of individuals and secure storage of sensitive data.
To qualify the quality of the data, we must take into account the difference between the machines that collect it. For example, in medical imaging, the quality of X-rays and MRIs will differ depending on the device. As for confidentiality, the collection of data does not pose a problem in hospitals, because it is doctors who collect it, thus avoiding confidentiality concerns.
The question of anonymization is just as important. If you are given an X-ray, it is difficult to make it completely anonymous. Thus in the United States, students managed to find the radios of a governor in a public database. Using an AI, they cross-referenced these x-rays with the age of the governor in question and managed to determine in which hospital the x-rays were taken. In France, all sensitive medical data must be hosted in secure data centers, labeled EDS, health data warehouses.
What is the future of AI and robots in medicine?
These technologies will continue to improve and become more popular. Their benefits are multiple:
- Address the lack of medical personnel. We can imagine the installation of terminals at the entrance to hospitals which will direct patients according to their symptoms. Or even teleconsultation booths will multiply and provide regular care or follow-ups (hypertension for example), thus fighting against medical deserts.
- Win time. Robots will be able to perform more surgical procedures, such as automatic catheter placement, and thus save medical teams precious time. Robots will gain autonomy during operations by having more precise gestures than a surgeon. Thanks to AI, voice recognition already makes it easier to enter medical reports and enter patient data via terminals and tablets. Contactless interfaces will develop with voice operations and medical records appearing in the doctor's augmented reality glasses.
- Personalize treatments. Thanks to AI, treatments will be much more finely tailored to each patient. Our digital medical record will follow us throughout life.
