Applications of Artificial Intelligence to Health

Silvia Stefanelli
Alice Giannini

It is undebated that nowadays the application of technologies based on Artificial Intelligence for medical use is spreading more and more. We have dealt with the basics of Artificial Intelligence here.Now, it’s time to get to the heart of this column: how is AI used in the healthcare sector? How can “machines” learn and apply the standard of medical ethics? Finally, will we really need doctors in the future? Spoiler alert: the answer is yes.

In 2018 the Wellcome trust, a foundation whose mission is to advance health through education and research, commissioned Future Advocacy to conduct a study on the ethical, social and political implications of the use of AI in health and medical research. The result was an outstanding report[1] which divided current and potential uses of AI in healthcare into 5 categories:

  1. Process optimisation: it refers to the use of AI to enhance “back-end” processes in healthcare such as procurement logistics and staff scheduling;
  2. Preclinical research (e.g. drug trials)
  3. Clinical pathways (e.g. diagnostics)
  4. Patient-facing applications (e.g. delivery of therapy or of information);
  5. Population-level applications (e.g. mapping of epidemics) 

We will adopt the same method of distinction throughout this article, providing the most relevant examples for each category.

Thus, before we begin, it is important to make a clear theoretical statement: what we see today are just examples of so-called “narrow” AI systems as opposed to “broad” AI systems. Narrow AIs are designed to execute specific and defined tasks. On the other hand, broad AIs should be capable of intelligent behaviour and to handle any intellectual task.

All AI-systems that are in the market today are  based on weak AI, and broad AI nowadays only belongs to science fiction. 


1. Process Optimisation

Corti is an AI designed to assist emergency dispatchers (human operators receiving emergency calls) and to oversee the whole dispatch process. It also analyses audio data from emergency calls (speech patterns, background noise, etc.) in order to detect cases of cardiac arrest.

2. Preclinical Research

AtomNet, a product of Atomwise, is an AI drug-discovery-software based on artificial neural networks (specifically, convolutional networks). During drug- discovery researchers need to identify the biological cause of the disease, which is usually a protein that will be targeted by the drug. In order for this to happen, the molecules belonging to the potential drug need to be able to bind to the disease’s protein and to produce a certain effect (i..e bioactivity).

AtomNet is able to analyze billions of chemical compounds to predict the bioactivity of small molecules. It then examines stimulations of how the potential medicine will affect human body.  As of today, AtomNet has been used to find a drug candidate to cure Ebola.

3. Clinical Pathways

Enylitic is a stand-alone software-as-medical-device based on an AI which interprets X-ray, CT and MRI scans at a speed of 15 milliseconds per image. It has been listed by the MIT[2] as the 5th smartest artificial intelligence company.

Freenome is an AI-powered blood test which combines molecular biology and advanced computational techniques to identify hidden patterns in the detection of cancer when it’s symptoms-free.

It aims at classifying billions of patters that define tumor types in order to help oncologist predict which therapy is likely to work best. The genomic cancer data collected will be used to reveal new targets for the next generation of cancer therapies.

4. Patient-facing applications

Buoy Health created a chatbot which helps the patient understand its symptoms. “By your side, the moment you feel sick” is one of the mottos displayed on their website and their chatbot is described as a “warm, comforting voice” which aims at letting the patient know that he’s not alone and at guiding him/her to the right care. This AI-based symptom and cure checker has been implemented, amongst others, by Harvard Medical School.

PatchAI is an AI-based chatbot designed to follow patients during clinical trials. Through PatchAI, doctors can monitor engagement, therapy adherence, symptoms, quality of life and other information which are directly given by the patients in real time through the Virtual Assistant.

Robot-Assisted Surgery: this general category includes virtual reality-enabled robotics for surgery. There are a variety of examples, such as Mazor Robotics, which is used for spinal operations thanks to its 3D tools and autonomous recognizion of anatomical features) .

5. Population-level applications

BlueDot is an AI-driven algorithm which conducts automated infectious disease surveillance by analysing news reports, animal and plants disease networks and official government proclamations. It was founded by a doctors specialized in infective diseases and it has lately been in the spotlight since it started sending warnings to its clients to avoid the Wuhan before the WHO alerted the public of the outbreak of the Corona virus.


The rapid spread of these new technologies in a sensitive environment such as health care comes with risks.  Some of the major concerns relate to the impact of Artificial Intelligence on patients' rights.

In the previous article we have already briefly highlighted some profiles of probable contrast: 

  • The right not to be subject to a decision based solely on automated treatment (right to human intervention): the GDPR (EU Reg. 2016/670, art. 22) guarantees the right of the data subject not to be subject to decisions based solely on automated processes if this decision produces legal effects on him/her or significantly affects his/her person;
  • The right to be informed about the logic used for automated decision making: Article 13 of the GDPR stipulates that where the holder uses an automated decision making process, the data subjects have the right to receive significant information about the logic used. In the case of artificial intelligence, quite a few problems arise, since the technology used is the value of the product itself and is therefore covered by confidentiality;
  •  The risks of data bias: artificial intelligence systems run the risk of reinforcing existing discrimination, on the one hand because they are "trained" on the basis of datasets that do not represent the whole population, and on the other hand because the algorithm in the course of the decision-making process could reflect the opinions of their developers;  
  • Accountability: who should be held responsible in case of an AI error?
  • Regulation: what is the European Union's approach to AI?
  • The future of the health care profession: will "machines" take the place of doctors?

The European Union has recently begun to address these issues.

In the next article we will give a general overview of the regulatory approach to Artificial Intelligence of the European Union.

Sources: Sam Daley, Surgical Robots, new medicines and better care: 32 exmaples of AI in Healthcare

[1] Find the full report here:

[2] See full article here:

This website uses technical cookies and third-party cookies to collect statistical information on users. To learn more and to manage cookies, click here. If you continue browsing through access to another area of the site or by selecting an item of the site (for example, an image or a link) you give your consent to the use of cookies and other profiling technologies used by this website. To hide this message, click here.