NexoVent: Artificial intelligence applied to the management of mechanical ventilation

Background: Mechanical ventilation (MV) is a widely used method of life support. Its proper management affects the outcome and depends on the expertise of the operators. However, there is a significant lack of knowledge on the part of most professionals, especially those who have recently graduated. Aim: To develop a platform based on computer vision and artificial intelligence capable of automatically evaluating and suggesting real-time adjustments to mechanical ventilator parameters. Methods: The NexoVent decision flow was created on the Miro platform and prototyped in Figma. App development used Flutter 3.10. For the computer vision algorithm, 200 monitor images of the Maquet® mechanical ventilators (Servo S and Servo I) were collected. Using Python and the OpenCV and TensorFlow libraries, the images were treated and the data extracted to a Firebase® database. The accuracy of computer vision was analyzed. Results: Data were extracted with an accuracy of 95.41% and precisely entered into a database. NexoVent uses this data to automatically analyze and inform the operator about any necessary adjustments for accurate and safe MV. The tool also guides the evaluation of measures such as Plateau Pressure, Pocc and calculation of Pmus and ∆PL,dyn, among others. Conclusion: NexoVent is an interface between the MV and the operator, capable of recognizing ventilation parameters and presenting the best adjustment options for each scenario. Computer vision showed excellent accuracy for capturing monitor data through a photo captured by smartphone. The data is successfully sent to a database. Thus, the best MV adjustment is guaranteed, avoiding common problems related to MV mismanagement, such as VILI.