Ventilator Control with Neural Networks and Fuzzy Logic: The Clinical Advisor Project
The Clinical Advisor is an automated system that continuously monitors a variety of information about the patient's respiratory system, calculates non-invasive estimates of important parameters (such as patient's effort) using complex neural network models and recommends settings (e.g. amount of pressure support, PEEP, IMV rate etc.) to the clinicians based on these predictions. The main goal of the system is to help clinicians make quick and correct decisions about respiratory support which will in turn improve patient care and reduce costs. The Clinical Advisor system can be integrated with ventilators or respiratory monitors to provide integrated respiratory solutions.
Currently for clinicians, the main hindrance to make efficient decisions or to react quickly to patient condition is information overload. Various ventilators, monitors and other devices provide so much information in various forms that many clinicians don’t have the time or background to interpret and properly react to the data. The main vision of Clinical advisor is to extract additional information from existing data and perform “data fusion” to provide important recommendations to the clinician.
The Clinical Advisor has three subsystems which focus on optimizing patient oxygenation, ventilation and effort. The main algorithms that are part of the clinical Advisor include 'detection of patient effort non-invasively', 'detection of intrinsic PEEP in patients receiving respiratory care' and 'maintaining proper oxygenation at all times'. The Clinical Advisor also includes a rule-based 'Fuzzy Logic System' that will provide advice on how to optimize the ventilator. The following figures give an overview of our data collection system and and the preliminary design of the Clinical Advisor user interface.
The Clinical Advisor will help clinicians make quick and correct decisions about respiratory support which will improve patient care and reduce costs. It is flexible enough to be used in different stages of respiratory care for different objectives and can be customized (in terms of parameters and/or rule base) at the hospital/departmental level. Currently we are performing clinical studies at several hospitals in the U.S. in order to further validate our algorithms and models. For more details about this project, please take a look at 'abstracts' of some of our published work or send us an email at email@example.com.