Alexa, is my cough normal? Refinement of a Machine Learning Algorithm Using Capsaicin Invoked Cough Signals for Cough Categorization

Respiratory diseases cause about 20% of deaths worldwide and are often under-diagnosed (Jose´ et al., 2014). Under-diagnosis implicates many health consequences such as increased respiratory complications and mortality, necessitating a higher rate of detection. Coughing is an important symptom of respiratory diseases and certain characteristics of coughs can indicate the presence of a respiratory condition. Previous studies have used Machine Learning Algorithms (MLA) and Artificial Intelligence to detect voice disorders and COVID-19 coughs. Given this success with MLA detection, training an MLA to detect characteristics of cough sounds can aid in diagnosis of respiratory conditions. To train this MLA, healthy voluntary and involuntary cough sounds will be collected. Appropriate refinement of an MLA requires high quality and appropriately diverse inputs to ensure accurate outputs and to avoid exacerbation of existing health disparities that can affect patient health outcomes. Appropriate data collection, pre-processing, and batch-normalization of data are vital processes to increase the accuracy and quality of MLA outputs. Thus, the obtained diverse cough sound data will be pre-processed for effective batch-normalization to ensure appropriateness and quality of the output. Involuntary coughs will be invoked by administering capsaicin, the pungent element of hot peppers. Capsaicin induces the cough reflex comparable to initiating a real cough. Refining the MLA with these invoked healthy coughs will ensure a more accurate distinction between healthy and disease related coughs. With the availability of an accessible tool, respiratory conditions may be easily screened in a variety of applications including in primary health, monitoring patient progress, and Telehealth use.