Recognition of Respiratory Dysfunctions Using Algorithm-Assisted Portable Airflow Sensors

Abstract

Respiratory diseases are becoming a severe health threat. To prevent exacerbation with early diagnosis, there is an urgent need for developing a respiratory function assay with ease of access. Tidal breathing pattern reflects a combination of the existing lung condition and the physiological demand. However, the interpretations of breath pattern remain underexplored. In this study, lung simulator with various pathological parameters was used to reconstruct the breath pattern of patients with chronic obstructive pulmonary disease (COPD) and interstitial lung disease (ILD). Breath pattern was recorded using two flow sensors. Three machine learning algorithms, including convolutional neural network (CNN), long short-term memory (LSTM) and support vector machine (SVM), were applied for disease identification. Results showed algorithmic analysis can achieve over 80% accuracy, and two levels of obstructive severity of COPD can be determined. With the assistance of algorithms, similar results can be obtained using a portable sensor. In contrast to the heavy professional and complex equipment requirement of the current methods, this proof-of concept method shows the potential of using a low-cost portable sensor for respiratory function monitoring. This approach can provide a basis for preliminary diagnosis, and may further contribute to point of care testing for respiratory health.