Monitoring and decision support system for artificial lung ventilation
Industrial enterprises have benefited from machine learning data processing for many years. Predictive maintenance helps to avoid breakdowns and reduce downtime. Augmented analytics allows to choose the optimal operating modes for equipment and reduces energy consumption.
To monitor patient’s condition, a large number of sensors are installed and special equipment displays data online on the device screens. Devices for centralized management of patient data are also exist. But as a rule, these solutions are quite expensive and have a number of limitations, in particular, they do not allow the use of predictive analytics methods to the collected data.
This is exactly the problem we faced when we began detecting asynchronisms during artificial ventilation using machine learning algorithms. We described this case in another material, and now I want to tell how we built a system for receiving, storing and outputting data from each ventilator online.
The clinic where we worked had Hamilton ventilators. They have RS-232 data interface, which allows to connect the data receiver directly to the device and get all information about the patient online.
Then the data was transmitted to the server, where it was further processed by various algorithms, including machine learning methods. The server also hosted auxiliary systems that check the completeness and continuity of the incoming data.
In a convenient and intuitive interface, designed in close cooperation with medical specialists, you can get all the necessary information about patients, and this can be done at any time from any device with Internet access: tablet, mobile phone, laptop.
The web application shows all current data, aggregated indicators for 3, 12, 24 hours, alerts are displayed separately.
For junior staff who have not yet received enough experience, the system displays recommendations on what to look for and what options for improving the patient's condition can help. The software does not take on the decision-making function, but by providing the data it helps to make the best decision.
In addition, data on asynchronisms are displayed on a separate screen, which allows to choose a more adequate mode of operation of the ventilator.
In an additional module, which was developed later, we made it possible for the attending physician to fill in data on the dynamics of the patient's health. When the treatment parameters and history of the condition are collected in one place, it is easier to draw up a plan for further treatment.
Since all data about all patients are collected anonymously and centrally in a database, specialists always have the opportunity to obtain historical data on a patient or a group of patients, to analyze treatment methods both manually and using automated tools. Currently, our AI department develops algorithms for predictive personalized analysis of patients' condition. We will definitely share our achievements with you in the near future.
It is also important that the data is completely depersonalized. The application does not store any information that would identify the patient's identity.