Exploration of Cough Recognition Technologies Grounded on Sensors and Artificial Intelligence

Abstract

AbstractArtificial intelligence is ruling all industrial sectors and has its hand on the medical and healthcare field too. Cough is a symptom of divergent respiratory disorder diseases from a common cold to the current coronavirus disease. Cough is not only extant in humans, but it similarly found to be existing in numerous animals primarily in pigs [1]. Cough is generally a good self-reaction of the body to prevent secretions and its blockages in the upper airway. The frequency, sequence and pattern of the cough reveal the disease along with its severity. Thus, sensing platform and artificial intelligence are used intensively for cough analysis. This chapter is to explore about cough detection and throws light on the various cough detection methodologies, the artificial intelligence algorithms implemented, features involved in cough detection and constraint existent in implementation. In architectural analysis of cough detection; divergent types of the sensors, auxiliary equipment and neural network sustenance instruments deployed are entailed. Cough detection is enacted by voluminous machine and deep learning algorithms using classifiers such as random forest, decision tree, logistic regression, support vector machine, feed forward artificial neural network, convolutional neural network hidden Markov model, multiclass classifier with multilayer perceptron model, and validation is achieved through K-cross validation. The chapter also articulates about the dataset availability of various patterns of cough, the visualizing of sound pattern in frequency and time domain. Further cough is found to have two set of features namely superordinate and subordinate sound features. Superordinate features include Mel-frequency cepstral significant, non-Gaussianity score, Shannon entropy, energy, zero intersection ratio, spectral centroid, spectral bandwidth and spectral roll-off. Subordinate feature covers cough sequence type and duration, bouts occurred in a sequence, cough sequence number in prescribed interval time. The chapter also includes extensive analysis of above feature sets of cough sound. Hence, cough detection using artificial intelligence helps doctors to diagnose early and at ease. At times, it also overcomes the misdiagnosis of the disorders. The chapter also discusses in detail about the various datasets used for cough detection. Finally, includes the constraint of deployment of cough detection that covers the challenges in computational cost, size, budget and ease of deployment with ubiquitous computing.KeywordsCough detectionSensorsMachine learningDeep learning algorithmsClassifiersSuperordinate featuresSubordinate features