In a series of articles byEconomist who try to imagine what the future will be like, entitled “What if?”, the imaginary winner of the Nobel Prize for Medicine of 2036 is YULYA, an artificial intelligence that has saved 4 million lives thanks to the overcoming of the antibiotic problem -resistance, the phenomenon whereby many bacteria resist the activity of antibiotics making them ineffective.
In general, artificial intelligence is the science that deals with how to create intelligent machines, and that has found in the possibilities offered by information technology the most practical way to obtain this result. The ultimate goal of artificial intelligence is to make software, which in turn have been called artificial intelligences, capable of achieving goals and solving problems as a human being would. Or rather, in some cases.
In the scenario imagined byEconomist, in 2034 YULYA – designed to develop new cancer treatments – is accidentally allowed access to a huge database of drug studies. In this way it is able to improve its knowledge on antibiotic resistance, finding new combinations of drugs and thus solving a problem currently considered by health authorities as a serious threat. It is obviously a fictional story but it tries to hypothesize what will happen in the future based on the scientific evidence currently available.
It is not a foolish suggestion to think that in the future artificial intelligence can cure people or help predict disease: in recent years, doctors and researchers around the world have collaborated and discussed how to exploit such a powerful technological resource. They also evaluated its limits, possible distorting effects and errors, which also in the health sector depend on how this technology is structured and how many and what data is used.
“Artificial intelligence can improve healthcare and medicine around the world, but only if ethics and human rights are put at the center of its design, implementation and use”, reads the report “Ethics and governance of the ‘artificial intelligence for health‘ published at the end of June by the World Health Organization (WHO) and the result of a two-year consultation between experts.
Although often perceived with a science fiction or dystopian meaning, artificial intelligence is already used in many disciplines of medicine as a support tool to improve judgment and in general the work of doctors. The main fields of application are diagnostics, the development of new drugs and vaccines, rehabilitation and telemedicine. It is useful in analyzing images for radiological diagnosis in oncology, especially for chest radiographs, colonoscopies, mammograms, brain CT scans; it is also used for non-radiological examinations, in dermatology and in ophthalmology, and helps to shorten the analysis time (sequencing) of RNA and DNA.
There are many possible developments: it could allow doctors to identify events of a certain severity such as stroke, breast cancer, coronary heart disease, pneumonia and diabetes early on. Even in the case of full-blown illness, artificial intelligence could be used to integrate patient medical records during the study of a cure or to obtain more detailed information before making delicate treatment decisions such as hospitalization or surgery.
Technology can also improve communication between healthcare facilities, family doctors and patients, and through constant monitoring help sick people have more awareness and control of their health conditions, of how and when to take drugs, of how important it is. have a healthy diet and regular physical activity.
– Read also: The day Siri tells us if we’re sick
The development of artificial intelligence in medicine is linked to the significant increase in data: at any time of the day all people produce a significant amount, and more and more detailed. A trivial and immediate example is the Health app on iOS, the operating system of iPhones, which counts the steps taken in a day and observes the heartbeat, if connected to a device such as a smartwatch. Even when you undergo an examination in the hospital, data is collected in different forms: they can be simple, such as blood pressure measurement, or complex like a latest generation CT scan, an in-depth investigation technique that allows you to reconstruct a three-dimensional image of the organs.
Data feeds several analytics systems that rely on algorithms and are essential to making AI truly intelligent. One of the most used analysis systems is the machine learning which consists in the study and construction of algorithms that allow computers to learn to perform a precise task such as recognizing an image or making reliable predictions, starting from a set of data provided by developers. The algorithm does not learn what a dog is based on its characteristics – it has four legs, two eyes, it can have ears of a certain size and a nose of a certain shape – but it can identify it after learning how it is done by observing thousands. of photos that the developers have categorized as dog images.
Understanding how the machine learning it is important to understand the potential of artificial intelligence in medicine, where data abound even if until a few years ago they were a little or badly used resource.
– Read also: How to teach a computer to learn
(Unsplash / Irwan Iwe)
Giuseppe Sollazzo is the “Head of Artificial Intelligence Skunkworks and deployment” of the British National Health Service (“The most hipster name in my professional life”, he jokes). It means he leads a team of data analysts who are tasked with figuring out how to use AI in the healthcare system and hospitals. The department of Sollazzo coordinates some experimental projects in hospitals to find possible applications of artificial intelligence in the health field, new methods of analysis, developments and improvements, errors and limitations, and in this way produce a model to be taken to all British hospitals.
The most promising area concerns image analysis. Artificial intelligence can observe the CT scan, a three-dimensional image, to understand how a disease evolves over time and therefore if a person’s condition improves or worsens. A doctor can understand this by looking at the CT scan thanks to studies and experience: through the machine learning, an artificial intelligence can analyze the same images more quickly, with a greater chance of detecting tiny lesions, and help the doctor make decisions.
Is there a way to predict how diseases like glaucoma (a chronic and progressive disease affecting the optic nerve) evolve before conditions become more severe? It is one of the questions we are asking ourselves and to which artificial intelligence can give an answer through the analysis of millions of images ”, explains Sollazzo.
One of the most relevant problems is the lack of detailed data to study models that are valid for all people. The Sollazzo department is working with a group of dermatologists engaged in the search for an effective and universal model for the treatment of pressure sores. “The impetus for this work came from the fact that there are few studies on the evolution of pressure ulcers in non-white patients: there are no accurate databases with images of black people,” he explains. “We are trying to understand how to create a platform to collect this data and create a predictive model of the evolution of the plagues that is not subject to bias.”
If artificial intelligence is trained only with images of sores from white people, the risk of errors in analyzing exams on a black person is very high. «One of the most interesting questions we are asking ourselves is whether a single model can be created for all people or whether a model must be studied for each degree of the phototype scale, ie skin color. These are the concrete consequences of the lack of data, ”says Sollazzo.
Remedying the scarcity of data is also complex because health data is delicate and cannot be collected and managed lightly. Precautions and safeguards are needed to avoid abuses such as improper use by companies. For some years also in Italy health and artificial intelligence experts have been comparing and studying possible applications in health care.
One of the most relevant examples is the “Citizen’s medical record” activated in 2012 by the autonomous province of Trento. It is an electronic health record that not only stores and analyzes the data collected in the hospital, such as the results of the exams, but also those produced every day by the patient.
A diabetic person, for example, enters blood glucose values, the type of meal eaten and what time they ate each day. “Artificial intelligence does not replace the doctor: the doctor uses artificial intelligence to monitor the patient’s health conditions more thoroughly and prescribe personalized treatments”, explains Paolo Traverso, director of the strategy and development department of the Bruno Kessler Foundation of Trento, which has collaborated in the development of the citizen’s medical record, and among the most qualified Italian experts in artificial intelligence. «The citizen’s medical record app interacts with people and doctors, and helps to better understand how a patient is and how his conditions evolve over time. Doctors bring knowledge, the app allows you to better observe reality: this is how the models of care are refined ».
At the moment the platform is used by over 140 thousand people in the autonomous province of Trento and the Bruno Kessler foundation has also entered into agreements with Emilia-Romagna, Umbria, Friuli Venezia Giulia and in hospitals such as Fatebenefratelli for which MumApp, a ‘app to follow the developmental stages of newborns in the first thousand days of life. “Artificial intelligence experts and also jurists work on all projects because the issue of data protection and informed consent is essential”, continues Traverso. “We used the ‘privacy by design’ method, an approach that considers privacy as central to development”.
– Read also: How Europe wants to regulate artificial intelligences
In the “Ethics and Governance of Artificial Intelligence for Health” report, WHO experts identified six principles to ensure that artificial intelligence for human care works in the public interest and with respect for privacy. The first is the protection of human autonomy: doctors must remain in control of health systems and medical decisions. Another relevant principle is the need to ensure transparency through the publication of all information on the design of an artificial intelligence system.
Information must be easily accessible and stimulate debate on the correct use of technology. Furthermore, inclusiveness and fairness must be guaranteed. WHO explains that artificial intelligence for health must be “designed to encourage the fairest use and access possible, regardless of age, sex, gender, income, race, ethnicity, sexual orientation, ability or other characteristics. protected by human rights codes “.