The doctor warned about the risks of decoding the tests through a neural network

Neural networks allow you to quickly get an explanation of the analysis results, but such answers can be inaccurate and lead to erroneous decisions. Artificial intelligence is capable of misinterpreting indicators, not taking into account the patient's characteristics, and both reassuring a person without reason and causing unnecessary anxiety. ""Invitro" internist Evgeny Kosmynin told "Izvestia" about the risks of decoding analyses using AI on September 26.

According to the expert, language models do not verify medical facts, but form an answer based on available data. Therefore, they may come up with plausible but erroneous conclusions, refer to non-existent norms, and misinterpret the results.

In addition, AI is able to confuse units of measurement that differ by dozens of times, and not take into account the specifics of specific research methods. For example, the results of analyses performed by different test systems may differ, but the neural network does not always warn about this.

Another problem is the lack of complete information about the patient. AI does not know the history of diseases, medications taken, surgeries performed, hereditary factors, age, gender, and other circumstances that may affect the assessment of results.

According to the doctor, the same indicator may have different meanings for different people. For correct interpretation, it is necessary to take into account the patient's complaints, the dynamics of changes, the results of other studies and the reference values of a particular laboratory.

Herd instinct: neural network will detect diseases of cows in the early stages

The development diagnoses the first signs of lameness, mastitis and bursitis, and also reduces treatment costs.

The main danger of decoding analyses through a neural network is related to the decisions that a person makes based on such answers. AI can create a false sense of security, causing the patient to postpone a visit to the doctor, or, conversely, cause unreasonable fear and lead to unnecessary examinations.

Self—medication is a separate risk. After consulting with the neural network, a person can independently start taking iron supplements, vitamin D or hormonal drugs, without taking into account possible contraindications and causes of deviations.

"A neural network is a convenient tool for working with data, but not for working with your health. She doesn't know your history, doesn't see you, isn't responsible, and can make big mistakes in numbers, units, and norms," Kosmynin said.

According to the expert, the results of the tests should be evaluated by a doctor who compares the indicators with symptoms, age, gender, chronic diseases and medications taken. If necessary, the specialist will prescribe additional studies or reanalysis.

Kristina Vysotskaya, a lawyer and member of the working group on criminal procedure under the Commissioner for the Protection of Entrepreneurs' Rights in Moscow, noted on August 19 that AI in medicine can only be used as an auxiliary tool, and responsibility for the decision remains with the doctor. According to the expert, the doctor has the right to consider the conclusions of AI as a second opinion, but cannot fully rely on the algorithm.