Quantitative Analysis of Capnography for the Diagnosis of Cardiopulmonary Diseases

Abstract

Capnograph is the waveform that measures the instantaneous concentration of Carbon newlinedioxide (CO2) in exhaled air. As the anatomical and physiological characteristics of newlinean individual are reflected in the capnograph, it can be utilized as a powerful tool for the newlinediagnosis of many cardiopulmonary disorders. A speed of sound-based capnographic newlinesensor is proposed. It works on the principle that the speed of sound traveling through a newlinegaseous medium is dependent on the concentration of each gas. Automated quantitative newlineanalysis of the acquired capnograph is done to diagnose diseases such as Chronic Obstructive newlinePulmonary Disease (COPD) and Congestive heart failure (CHF). Three classification newlinetechniques namely, attention-based Convolutional Neural Network (CNN)- newlineSupport Vector Machine (SVM), Second Generation CNN and Deformable CNN are newlineproposed. The attention network weighs the features extracted by CNN in accordance newlinewith its significance and is given to SVM for classification. Second Generation CNN newlineincorporates the advantages of the lifting scheme to CNN. Deformable CNN allows newlineformless deformation of sampling in convolution and pooling layers of CNN. The classification newlinewas performed on the signal acquired using the speed of sound-based sensor newlineas well as the signal in the capnobase dataset. Moreover, clinical validation was done newlinefor the experimental results. The accuracy of 91.32%, 91.44% and 92.9% are achieved newlinefor CNN-SVM with attention, Second Generation CNN and Deformable CNN respectively newlinein case of signal from Capnobase. The accuracy for the signal from the speed of newlinesound based sensor is 90.68%, 90.48% and 92.16% for CNN-SVM with attention, Second newlineGeneration CNN and Deformable CNN respectively. The proposed non-invasive, newlinelow cost and portable capnographic sensor can broaden the application of capnograph newlinefrom intensive care units to primary healthcare. Improved accuracy and reduced computation newlinetime achieved indicate the ability of the proposed classification techniques to newlinediagnose cardiopulmonary disorders in real-time.