All news

Will artificial intelligence be able to replace a doctor?

11.08.2021
Topics

Since the invention of machine learning technologies, there has been a constant dispute - whether artificial intelligence can replace a number of professions, or not. And everyone is especially concerned about the fate of socially significant ones, such as a teacher or a doctor. Some are convinced that these professions will change beyond recognition or even disappear, others — that they will remain, and in the same form.

Our company "Medical Screening Systems" (Celsus) has been developing and implementing AI solutions for analyzing medical images for more than two years. During our work, we have sometimes met experts, doctors, and patients with absolutely different points of view — both about machine learning technologies, and whether they can replace a doctor. And since we have significant competencies and experience in this field, we decided to share our point of view about this.

We should say that we do not believe in replacing a doctor with artificial intelligence. But we believe in improving the quality of diagnosis and treatment of diseases thanks to the collaboration of the medical community and developers of AI systems.

Argument #1: A doctor without AI is still a doctor. AI without a doctor is nothing.

Highly qualified doctors take part in the training of medical artificial intelligence and in the evaluation of its work. At least, this is what happens if developers who want to create a product that is safe and applicable in clinical practice are experienced. For example, radiologists were involved in the training of our service "Celsus" from the very beginning - for consultations and data marking. Each image used in the training was independently marked by several specialists and in case of divergence, the image was given for additional examination.

The process of training and developing the system is endless. And not only because of our perfectionism but also because of new tasks and changing conditions. A recent example is a COVID-19 pandemic, which suddenly made it necessary to do a huge number of CT examinations of the lungs. Developers of artificial intelligence systems (including us) quickly began to train their models to detect signs of a new coronavirus infection on scans. And again, the doctors were very important - they helped us collect data for training (including examinations with confirmed cases of COVID-19), they marked the signs of viral pneumonia on the images.

In addition, the pandemic is a pandemic — and lung cancer, for example, has not gone away, and it is also important not to miss it on CT examinations (even when we check for covid). The problem is that artificial intelligence can only perform a specific task — that is called "who studied and what studied". This means that we again needed to retrain the system, combining two directions in it — so that it could check the images for signs of both diseases.

And to solve this huge problem, we again attracted the best radiologists. At the moment, our solution for detecting COVID-19 and lung cancer has already reached the calibration testing stage. We hope that very soon it will help doctors in real clinical practice.

Argument #2: AI is not a medical "megamind". This is a tool for solving routine tasks.

To reveal the essence of this argument, we need to answer an important question: why do developers create medical artificial intelligence at all? Let's go through the list of possible answers. Insure and complement the doctor - yes.  Take on part of the routine work, to save time for a more detailed examination of each case - yes. And where is the option "replace the doctor"? Why isn't it on the list? It is because none of the developers has such a goal.

Those artificial intelligence systems that are being created and developed are just tools that, in the right hands (=the hands of a doctor), can optimize the work processes of medical institutions and improve the quality of medical services provided. No one is working on creating an artificial "super doctor" who would analyze the images, communicate with clients, study the anamnesis up to the third generation, and prescribe treatment. At the moment, there is neither such an opportunity nor such a goal.

Much more important to us seems to be a careful, systematic solution of less ambitious, but more realistic tasks. For example, in the passing 2020 year, we finished pilot projects on the use of our service "Celsus.Mammography" in three regions of the Russian Federation — the Republic of Dagestan, Tambov, and Bryansk regions. And it gave concrete results. First, the use of AI has reduced the time spent by a doctor on analyzing examinations by a third. Secondly, there were 29 cases when artificial intelligence detected pathologies in the images that were invisible to the radiologist's eye, and additional examinations confirmed the presence of breast cancer at an early stage in the patients.

Argument #3: Medicine means a huge responsibility. Only a person can take responsibility.

Artificial intelligence has been widely and successfully used in various spheres of public life for several years — it ranks the news on Instagram based on our individual interests, predicts the weather, and even evaluates the creditworthiness of potential bank borrowers. But with its use in medicine, everything is much harder. Mainly because of the high risks that invariably accompany medical activities. The moderate conservatism of the medical community is quite justified — as well as the need for an assignment of responsibility

The most accurate description of artificial intelligence systems like ours is "medical decision support system". We are sure that in the future they will be used as a kind of " supplement to the doctor", helping him not to lose sight of important factors and make a more balanced decision. But only a doctor can make the final decision when it comes to a person's life and health.

Argument #4: A person needs a person.

And finally, artificial intelligence cannot replace live human communication between a doctor and a patient. And it not only gives the patient an explanation about his condition and a favorable attitude to treatment. Sometimes it helps to solve real medical riddles. Perhaps, every experienced doctor will have a couple of stories about how, with the help of a confidential conversation, he managed (often not even on purpose) to solve a non — trivial medical problem - for example, to identify an unfavorable factor in the patient's daily environment that provokes the development of the disease. Artificial intelligence is focused only on solving specific tasks, it cannot communicate.

Conclusion.

Artificial intelligence is a promising technology that is already being implemented into medical practice and is proving its effectiveness. There is no doubt that in the future it will be used in all (or almost all) medical institutions, improving the quality of diagnosis, treatment, and prevention of various diseases.

Will artificial intelligence replace a doctor? No, it will not replace it (at least in the context of the next decades). But, a doctor who uses the features of artificial intelligence in his work will replace a doctor who does not use them.