Decision Support Systems (DSS) are gaining popularity in modern medicine. The main purpose of such computer systems is to help people who make decisions in difficult conditions for a complete and objective analysis of objective activity.
Medical DSSs in the articles are called CDSS, adding the word "clinical" and receiving the Clinical Decision Support System - Clinical decision support system. These systems are aimed, first of all, at assisting in the appointment of procedures and drugs in the treatment and rehabilitation of patients.
In medical processes, a fairly large amount of action is strictly limited by the clinical protocol. Understand
The picture shows a fragment of the clinical protocol from the official document of the Ministry of Health of the Russian Federation "Clinical guidelines: multiple sclerosis" with the interesting name "Algorithm for managing a patient with exacerbation of MS."
If someone intuitively understands what needs to be done in the case when there is an effect after using plasmapheresis, then he definitely knows something about this life and not only. That is, the main problem with these protocols is that almost every arrow has a question “what if not so?”. Due to this problem, it is not possible to use clinical protocols in their raw form without the direct participation of a highly qualified specialist.
It is clear that it is possible to increase the efficiency of medical processes if, for the sake of interpreting strictly regulated processes, you do not have to tug on the doctor every time and, for example, shift the dispensing of medicines to junior medical workers who will be helped by the DSS. Just in case, I will explain: in many clinical processes, even the choice of drugs strictly depends on the numerical results of the tests, and the doctor cannot simply take and prescribe something else, simply because he wants to.
Now, knowing the scale of the problem, let us return to automata models. This post is designed to show why the use of automata models in the development of DSS is actually not only justified, but also intuitively logical.
Let's start with the simplest: the automaton has many states and many transitions, and the patient also has a state. In the previous sentence, the words "state" are nothing more than homonyms. Moreover, a person cannot be both alive and dead at the same time, which means that, as in the case of an automaton, a person has exactly one state at a time. Why not try to compare them to each other?
Since we are going to use the automaton model in the DSS, and even for the implementation of the clinical protocol, it makes sense to use Deterministic Finite Automata. Let's take a look at the properties of such a model and make sure that they are very similar to the required properties of the required system.
As it turned out, it is impossible to simply google "properties of a deterministic finite automaton", since no one formulates them explicitly - perhaps because these properties are too obvious, perhaps not interesting to anyone. Nevertheless, we highlight two that are quite useful for the topic under discussion:
- Conditions for transitions from each state are physically orthogonal.
- The logical disjunction functions describing such transitions are identically equal to one.
Let's see what it gives us. In the case of the first condition, the patient, for example, cannot simultaneously be in different stages of exacerbation. Or it is impossible, for example, to give the patient medicine and not give it at the same time. The second property guarantees that the entire range of actions proposed in the current state covers the set of all options completely, which will help to avoid the situation described earlier about "there is an effect after the application of plasmapheresis", that is, regardless of the input sequence of the machine / work with the patient, we we will always know the next state.
Well, the final obvious advantage of using the automaton model is, of course, a huge variety of verification tools, which, in turn, when used correctly, guarantee that if the automaton model is composed correctly, then after integration into the system it will always work correctly.
In conclusion, I would like to note that, despite the described prerequisites, the integration of an automaton model is not as simple a process as we would like. For example, unusual cases of the course of the disease are not always described in the clinical protocol and require mental activity on the part of a specialist, but all "stupid" things can be left for DSS with a clear conscience!