Quantitative and Real‐Time Evaluation of Human Respiration Signals with a Shape‐Conformal Wireless Sensing System

Abstract Respiration signals reflect many underlying health conditions, including cardiopulmonary functions, autonomic disorders and respiratory distress, therefore continuous measurement of respiration is needed in various cases. Unfortunately, there is still a lack of effective portable electronic devices that meet the demands for medical and daily respiration monitoring. This work showcases a soft, wireless, and non‐invasive device for quantitative and real‐time evaluation of human respiration. This device simultaneously captures respiration and temperature signatures using customized capacitive and resistive sensors, encapsulated by a breathable layer, and does not limit the user's daily life. Further a machine learning‐based respiration classification algorithm with a set of carefully studied features as inputs is proposed and it is deployed into mobile clients. The body status of users, such as being quiet, active and coughing, can be accurately recognized by the algorithm and displayed on clients. Moreover, multiple devices can be linked to a server network to monitor a group of users and provide each user with the statistical duration of physiological activities, coughing alerts, and body health advice. With these devices, individual and group respiratory health status can be quantitatively collected, analyzed, and stored for daily physiological signal detections as well as medical assistance.


Supplementary Information
Note S1.

Note S2. The theoretical working model of capacitive sensor
Due to the parallel relationship between initial capacitance and additive resistance, the total impendence Z under passing gas molecules can be expressed as

So we can find
We can thus obtain an imaginary part (capacitive value) as The above equation can be simplified to Thus obtain an updated capacitive value as Finally, the normalized capacitance change for the device can be calculated as We can see that relative the capacitance fluctuates along with the parallel resistance, induced by the human respiration process.

Note S3. Temperature calibration of the respiration sensor
For the 100-nm thick respiration sensor, as shown in Fig.2g, the capacitance errors are linear between 25-50 °C (T). The error is calculated to be c.a. 0.01% in this range. If we define the initial capacitance value as C 0 , the changed capacitance C should be )) That is ) While T can be calculated according to the resistance output, thus the corrected output C x should be ) In most occasions, such correction is not necessary especially for the physiological temperature range. Nevertheless, all the temperature data would be saved to the memory for the assisted respiration analysis. This method would be useful in other potential applications under extreme temperature conditions.

Note S4. The model validation parameters
The sensitivity (True Positive Rate) refers to the probability of a positive test, conditioned on truly being positive. Thus 9 true subjects among 10 detected cough patients means a 9/10 sensitivity. The Specificity (True Negative Rate) refers to the probability of a negative test, conditioned on truly being negative. Thus 37 true subjects among 40 detected non-cough patients means a 37/40 sensitivity.   although baseline-shift occurs on the sensing curve of raw data from c.a. 23 pF to c.a. 29 pF induced by large sensor deformations, the following first step of data processing would effectively correct this factor to ensure the correct classification of respiration signals using our ML algorithm. That is to say, the respiration sensor would be more sensitive to the relative changes rather than shape deformation. c: as for the coughing detection, we take the sensor with bending deformation in a mask as an example, the processed data would also be classified into coughing status. This result can also be found in our supplementary video S2 (the rest status mixed with coughing), where the sensors are bent to fit the shape of mask.) reduce the abnormal humidity influence. One is to assemble a breathable TPU layer upon sensor substrates (The substrate is more easily affected by humidity due to its hydrophilic properties after oxygen treatment during deposition). The hydrophobic TPU would effectively protect the sensor and reduce the sensing bias shift after wearing them for a period of time.
The other one is improving by our ML algorithmic model and data processing. The derivation process would focus more on the respiration trend and relative value gain despite of some inevitable bias shift caused by abnormal humidity conditions.) Fig. S13. The cost details of our sensing system. (Although the sensors have good stabilities benefitted by its non-invasive design, we do not recommend continuous use for more than a week which might may cause bacterial growth. Of course, cross-use between different users is also not recommended.)