Can a machine cure a human? To a certain degree yes — robots have been used in surgeries for more than 20 years. Almost all leading digital corporations are developing smart medical products, services, and processes. In total, according to the research firm Venture Scanner, such developments are being conducted by more than 800 companies worldwide.
The field of robotics is one of the most important applied sciences. It helps create high-precision machines for complex medical tasks. This process rests on hours of learning of intelligent systems, the experience of qualified developers and the work of doctors.
It would not be farfetched to say that AI is one of the most promising directions in the medical sector. In 2016, healthcare AI projects attracted more investment than AI projects within any other sector of the global economy.
Why do we need a metal doctor?
Artificial intelligence In medicine can improve the efficiency of diagnostics due to its ability to work with big data. There is a known case when the smart diagnostic service by IBM Watson discovered a rare form of leukemia in a 60-year-old patient with an incorrect diagnosis. To do this, the system “studied” 20 million scientific articles on cancer in 10 minutes.
AI reduces the amount of daily routine tasks that medical professionals are forced to solve. In addition, it is able to minimize possible errors. What is more, it opens up opportunities for the emergence of new professions related to the maintenance of medical digital systems. These systems are able to learn from each individual case and within minutes can be exposed to more cases than a clinician could read through in many lifetimes.
For example, Semantic Hub created an artificial intelligence-based service to automate the assessment of the drug potential before medical products are released into the market. The system collects and analyzes millions of documents, including scientific publications related to the disease, the purpose, and effect of the drug under development. Then the robot analyzes the information and makes a conclusion about the potential of the drug, taking into account all risk factors and competitive advantages. Previously, drug developers were able to study “manually” only 1% of such documents.
This is one of the reasons why an AI-based application can out-perform dermatologists at correctly identifying suspicious skin lesions. AI is being more and more trusted with tasks where experts often disagree, such as classifying pulmonary tuberculosis on chest radiographs.
AI capabilities are not limited to these examples only. Governments allocate a lot of money for research in the field of healthcare, including the development and deployment of smart systems.
However, among the excitement, there is a fair share of scepticism, with some urging caution at inflated expectations.
Why is it all not roses?
Many experts predict a rapid increase in the market for artificial intelligence — about 35% annually. According to BIS Research, by 2025 the total AI market in healthcare will reach $28 billion.
But not everything is so perfect. Obstacles arise in the implementation of AI technologies in medicine. And even medical specialists themselves sometimes question this technology, which causes reasonable distrust among them.
What causes the problems? Let’s try to figure it out.
This, of course, is about the quality and quantity of medical information. The data accumulated in patients’ medical records may be incomplete, contain errors, inaccuracies, and non-standard terms. Records do not have enough information about the patient’s life, habits, and behavior. Effective mechanisms for collecting this information do not simply exist so far.
Results of the analysis based on such data will always cause reasonable skepticism and attempts to improve its quality get stuck on the complexity of the process.
To eliminate this problem, learning options for artificial intelligence with the use of small amounts of information are now on offer. Successful examples of such learning are observed when we use a smartphone keyboard, where the system remembers and analyzes the words entered earlier and can predict the content of texts we are inputting. Face and music recognition applications are based on similar technologies.
With the successful implementation in medicine, the machine learning system is able to solve many problems — check drug compatibility or deliver diagnoses based on genetic analysis.
Smart machines — smart approach!
In addition to the algorithm itself, which can perform analysis with a high degree of accuracy, the success of innovations requires a successful project team. Such projects in medicine depend on how productively the participants interact.
The team should include specialists with a wide range of competencies in the domain area, mathematical algorithms, and approaches to information protection, programming skills, and visual data presentation. It is highly desirable that the participants possess not one but several different competencies in order to understand and complete each other.
The second limitation of the proposed innovations in medicine is the lack of transparency in the decision-making process when it comes to the intellectual core of the system. AI works using the black box principle. If there is an error in the algorithm, and the system has made the wrong decision, then it will be extremely difficult to find out why.
Nowadays IT companies are developing machines that can unveil the reasons for their decisions. American scientists are approaching the release of a similar product to the market. In particular, the Defense Advanced Research Projects Agency (DARPA) supports 13 research groups that are focused on solving this problem.
The complexity of approaches to data processing based on artificial intelligence poses another problem. We need to select and train personnel capable of efficiently using and maintaining systems with non-trivial algorithms.
Price and value
Emerging difficulties increase the cost of development, implementation, and introduction of solutions based on artificial intelligence. The high cost of projects is also associated with the need to configure a new system for the data accumulated in a particular medical institution, and to construct a qualified and motivated team.
This, in turn, casts doubt on the possibility of rapidly scaling up the technologies offered by startups. Scaling is possible, for example, in the case of processing medical images of the same type, but needs go far beyond these limits.
Industry experts agree: in the short term, the introduction of artificial intelligence will not lead to a significant reduction in costs. We must continue to look for areas where the use of artificial intelligence technologies will bring higher value.
Lord, protect us from hackers
We must not forget that in order to ensure the functioning of artificial intelligence, it is necessary to provide access to high-performance computing capacities, which medical institutions often lack. Accordingly, data arrays will have to be displayed outside the perimeter of the institution, and this threatens the safety of storage, which should be the first priority. It is no coincidence that many artificial intelligence implementation projects were stopped due to risks related specifically to information security.
One of the vivid examples was when after the start of successful collaboration, the US Department of Veterans Affairs terminated the contract with the startup Flow Health, a developer of an intelligent disease diagnosis system. As the United States Department of Health explained, the contract was nulled when it became known that the system processes confidential data. The department considered this a violation of the security of personal data, although no proven leaks were found.
The expert community is fully aware of the existing problems and is trying to respond to them: specialists think through the principles of designing robotic systems, propose to discuss ethics of the use of AI in practice, and develop new options for introducing technologies. But let’s say it bluntly: it takes time to adequately work out approaches and standards.
Most of the issues that slow down the implementation of AI-based solutions in medicine are not at all related to the technological state of affairs. Most often these are ethical and administrative process problems, difficulties in collecting and processing data (e. g. providing the doctor with the time necessary to enter information into the database), problems of responsibilities, understanding mechanisms of artificial systems, etc.
What are the prospects then?
Will intelligent robots replace doctors in the future? It is still a difficult question to answer. Most medical startups are aimed only at simplifying and optimizing the medical flow process, and they do not provide for the replacement of a doctor with a robot. In addition, machine learning takes time, because we need to teach the neural network to “see” the problem in the image or to operate as a doctor.
Despite all the difficulties, AI-driven medical projects have prospects. We believe that in the next couple of years, artificial intelligence will be able to find its followers among pharmaceutical companies. It will be of good use in the search for new molecules and biological targets, in the virtualization of preclinical trials, and then in the analysis of clinical research data. When working with large companies, it is often possible to provide all the conditions necessary for a successful project launch.
All issues can be resolved. The main thing is that with the use of new approaches we should be able to move forward, and not cause greater harm.