Lung health diagnosis through cough sound analysis

[摘要] ENGLISH ABSTRACT: This study investigates a simple and easily applied tool for TB screening basedon the analysis of cough audio and objective clinical measurements.Tuberculosis is one of the most lethal diseases worldwide. There are variousdiagnosis methods for TB. However, in lower income areas, clinics lack fundsto afford expensive equipment and employ the trained experts needed to interpretresults.A database of cough audio recordings and clinical measurements was collectedfor this study. An automatic annotation system was developed using hiddenMarkov models (HMMs). The frame-accuracy of the annotation system is87:16%.For audio based classification we considered logistic regression and Gaussianmixture models (GMMs). We found that filterbank energy features outperformedMFCC features when used for audio classification, which could indicatethat cough audio contains information relevant to TB diagnosis that isnot perceivable by the human auditory system. Feature selection was used toinvestigate the importance of different frequency bands for classification and,it was found that the optimal results were achieved when combining featuresfrom the human vowel range (below 1000Hz) with features from high frequencyranges.As the main metric of evaluation, we used the area under the receiver operatorcharacteristic curve (AUC). This metric was chosen because it is not affectedby class imbalance in the dataset. Our best reported AUC was 94:94%, witha standard deviation of 4:62%, which was obtained using a set of just 5 filterbankenergies. We also showed that audio based classification obtains a higherAUC than classification on objective clinical measurements (meta data).Finally, we found that combining the audio and meta data classifier resultsusing classifier fusion improved how well the model generalizes. By combiningthe best audio classifier with the best meta data classifier, we obtained a sensitivity,specificity, accuracy, AUC and kappa of 82:35%; 80:95%; 81:58%; 94:34%and 0:6867 respectively.

[发布机构] Stellenbosch University