Pressure prediction for the personalized and automatic fitting of respiratory masks.

Abstract

Respiratory masks are important protection equipment for healthcare workers. User's discomfort due to a poor fit of respiratory masks over a very long period is a serious concern. The objective of this study is to predict the user's fit through prediction of mask's facial pressure on a mobile device. Personalized 3D printed masks were designed based on 60 face geometries. 3D scans and finite element analysis (FEA) results were used to develop machine learning (ML) algorithms for predicting pressures at the face-mask interface. Random Forest Regressor, Decision Tree Regressor, and Elastic Net were tested after standardizing input data. Predictions were made for 15 levels of mask tightening using linear force. Error indicators as Mean Absolute Percentage Error (MAPE), Median Absolute Error (MAE) and Root Mean Square Error (RMSE) were assessed, and the predicted mesh was calibrated against the FEA model using the Iterative Closest Point (ICP) algorithm. The models demonstrated their feasibility in reproducing the FEA results, with Random Forest Regressor providing the best pressure prediction (1.875 RMSE, 0.169 MAPE) and convincing 3D mesh results, respecting a 5% tolerance threshold. ML models were shown as feasible surrogates to FEA for eventual use on a mobile device.