Performance Evaluation of Wearable Nasal Sensor for Respiratory Assessments Using a NOVEL EDRS-LSTM Deep Learning Model

Recent advances in the wearable sensing technology haves paved the development of various wearable respiratory sensors that aids in monitoring the breathing patterns as well as to handle covid-19 like extraordinary situations. This specific research work highlights the performance of the wearable nasal mask integrated with multivariable sensors for breathing/respiratory assessments. A flexible miniaturized nasal sensor with electronics to measure the variation of left and right nostril has been designed and was attached on the inner part of the n95 mask for continuous recordings of respiratory signals. Three variables, temperature, pressure and humidity which were the integral part of the nasal sensor were then used to collect the breathing patterns with normal and deep breathing as well as breathing changes while walking. Inorder to assess the efficiency of the wearable nasal mask, a novel Encoder-Decoder Respiratory Signal-LSTM (EDRS-LSTM) deep learning model was deployed to perform multi pattern recognition. The proposed study involves 42 healthy volunteers with 35 minutes recording and a total of 2 trials was conducted. Experimental results showed an overall accuracy of 99.69% with AUC value 1.0 and F1score of 0.99 with the proposed deep learning model. The designed wearable with multiparameter variables with left and right nostril ensures the recognition of breathing patterns with an overall accuracy of 99.69% while comparing its performance with other known deep learning models such as RNN.