Speed of Sound-Based Capnographic Sensor With Second-Generation CNN for Automated Classification of Cardiorespiratory Abnormalities

Abstract

The measurement of carbon dioxide (CO2) concentration in exhaled air can be used as a diagnostic tool for many pulmonary and respiratory diseases. Capnography, the waveform that tracks the amount of CO2 in exhaled air, reflects the physiological and anatomical conditions of the human body. In this paper, we propose a novel speed of sound-based capnographic sensing mechanism for measuring the CO2 concentration in the exhaled air. The sensor response is further classified using our proposed second-generation convolutional neural network with SVM architecture. Automated quantitative analysis to differentiate between healthy subjects and patients with chronic obstructive pulmonary disease (COPD), between healthy and congestive heart failure (CHF), and between COPD and CHF has been implemented. The classification was performed on both the signal acquired using the developed hardware and the signal from the capnobase dataset. Furthermore, the experimental results were clinically validated. The proposed method yields an accuracy of 95.16%, 87.55%, and 90.19% for COPD/Healthy, CHF/Healthy, and COPD/CHF classifications, respectively. The development of a non-invasive, low cost, and small size sensor for capnography can extend the use of capnography from intubated patients to pre-hospital applications. Lower computation time and better accuracy achieved depict the effectiveness of the proposed method to be used in real-time diagnosis of cardiorespiratory diseases.