Computational Modeling and Data-Enabled Techniques Applied to Auscultation

Modern auscultation began with the invention of the stethoscope by Laennec in 1816 and auscultation with a stethoscope remains the mainstay of bedside clinical diagnosis. Although auscultation is a non-invasive and in-expensive diagnostic technique, the sensitivity and specificity of auscultation has however been limited by incomplete understanding between cause-and-effect, human-in-the-loop, dated measurement technique, contamination of sound by noise, and wide array of characteristics (timing, intensity, location, duration, configuration, pitch and quality) associated with sounds that need to be identified in order to make the diagnosis.

Clinicians and engineers have sought to transform auscultation into more objective and measurable modality that could be compared to disease phenotype and severity to enhance the ability of clinicians to screen for and monitor disease. The last decade has seen a paradigm shift from the traditional stethoscopes towards electronic devices that do more than just amplify sounds. These instruments that are sensitive to a programmable sound spectrum, offer abilities to record sounds and incorporate noise reduction algorithms. Along with that, there have been rapid developments in high-performance computing, data-driven science, and machine learning algorithms. These computational techniques combined with the modern electric stethoscopes can potentially allow the auscultation to become a reliable tool for diagnosis.

The current Research Topic is focused on those computational techniques applied to auscultation to mitigate the limitations, including but not limited to identification of source mechanism, automated signal processing (segmentation, noise reduction, etc), feature detection and extraction for disease conditions, diagnosis using the machine learning techniques, and computer model development.