Efficient and Explainable Deep Neural Networks for Airway Symptom Detection in Support of Wearable Health Technology

Abstract

Mobile health wearables are often embedded with small processors for signal acquisition and analysis. These embedded wearable systems are, however, limited with low available memory and computational power. Advances in machine learning, especially deep neural networks (DNNs), have been adopted for efficient and intelligent applications to overcome constrained computational environments. Herein, evolutionary algorithms are used to find novel DNNs that are accurate in classifying airway symptoms while allowing wearable deployment. As opposed to typical microphone‐acoustic signals, mechano‐acoustic data signals, which did not contain identifiable speech information for better privacy protection, are acquired from laboratory‐generated and publicly available datasets. The optimized DNNs had a low model file size of less than 150 kB and predicted airway symptoms of interest with 81.49% accuracy on unseen data. By performing explainable AI techniques, namely occlusion experiments and class activation maps, mel‐frequency bands up to 8,000 Hz are found as the most important feature for the classification. It is further found that DNN decisions are consistently relying on these specific features, fostering trust and transparency of the proposed DNNs. The proposed efficient and explainable DNN is expected to support edge computing on mechano‐acoustic sensing wearables for remote, long‐term monitoring of airway symptoms.

Wearable Neck Surface Accelerometers for Occupational Vocal Health Monitoring: Instrument and Analysis Validation Study

“…An in-house mobile app is also in development with features of NSA data visualization and vocal health feedback. We have already developed machine learning algorithms that are lean and efficient enough to classify upper airway symptoms such as cough and throat clearing on the NSA board [ 5 ]. The aforesaid system upgrades will broaden the NSA functionality to be more interactive and suitable for all-day monitoring.…”

“…A signal-to-noise ratio of 45 dB was achieved using the recorder’s multiple modes for background noise suppression. Further details and verification tests of the NSA system were reported in our previous publications [ 4 , 5 , 11 , 29 ].…”

“…Neck surface accelerometers (NSAs), a type of mechano-acoustic sensor, have been adopted as mobile health (mHealth) wearables for voice and upper airway health monitoring [1][2][3][4][5]. The vocal folds, which are housed in the larynx, oscillate at high frequencies (>100 Hz) when we speak or sing.…”

Deep Neural Network Regression‐Assisted Pressure Sensor for Decoupling Thermal Variations at Different Operating Temperatures

“…Among the various data‐driven models, deep neural network (DNN)‐based models have the advantage of demonstrating complex nonlinear relationships between the input and output parameters. Thus, DNN‐based methods have been widely used to reveal relationships in various applications, such as multipoint contact estimation, [ 31 ] cough prediction, [ 32 ] thermal mapping, [ 33 ] cancer screening, [ 34 ] nanocomposite property prediction, [ 35 ] inverse design, [ 36 ] and wearable electronics. [ 37 ] However, the thermal effect decoupling of the sensor signal distorted by the material and structure‐related complex effects has not been reported.…”