Cough Benchmarks

Chronic cough is a common condition globally, affecting 7.9% of the general adult population. It can lead to frequent consultations with the doctor and financial costs to the health service and the individual. Monitoring respiratory disorders, especially coughs, in ambulatory settings typically depends on self-reported assessments collected through highly subjective questionnaires by the practitioners. The quantification of the severity of these respiratory events is then biased by the patients, who often lessen the actual frequency of events, preventing the detection of signs of exacerbations of chronic respiratory diseases.

While efforts are being made to develop wearables to automatically detect and quantify cough events (typically based on recorded audio signals), such monitoring devices have not yet been incorporated into routine clinical practice. One of the main reasons is that the literature shows considerable discrepancies in the reported results between different teams, even while using similar hardware/data. 

This project builds on ORD datasets, community guidelines, and standards to propose a unified framework for validating cough event detection algorithms. The main objective is the development of standards that will unify the workflow for the validation of respiratory event detection algorithms to ensure data adheres to the principles of Findable, Accessible, Interpretable, and Reusable data. 

We will be contributing to Open Research Data in different ways:

● We will establish clear operational definitions of cough events based on updated definitions from field specialists. This necessary step will be the base for establishing robust labeling of data.
● We will standardize the data format for future open datasets by proposing a standard for the input modalities and the output annotation format. We will curate our previously acquired dataset according to these standards as a first step towards an open centralized platform.
● We will standardize the methodology to train and evaluate cough events detection algorithms by proposing a common strategy for training along with metrics to evaluate them.
Furthermore, we will develop software tools to split training and testing data and libraries that compute the performance metrics.
● We will then centralize all this knowledge, our dataset, standards, and tools on a website that will address common challenges both from ML and the medical community. 

cough

Related Publications

How to Count Coughs: An Event-Based Framework for Evaluating Automatic Cough Detection Algorithm Performance
Lara Orlandic, Jonathan Dan, Jerome Thevenot, Tomas Teijeiro, Alain Sauty, David Atienza
2024	arXiv