Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
149 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Cough Against COVID: Evidence of COVID-19 Signature in Cough Sounds (2009.08790v2)

Published 17 Sep 2020 in cs.SD, cs.LG, and eess.AS

Abstract: Testing capacity for COVID-19 remains a challenge globally due to the lack of adequate supplies, trained personnel, and sample-processing equipment. These problems are even more acute in rural and underdeveloped regions. We demonstrate that solicited-cough sounds collected over a phone, when analysed by our AI model, have statistically significant signal indicative of COVID-19 status (AUC 0.72, t-test,p <0.01,95% CI 0.61-0.83). This holds true for asymptomatic patients as well. Towards this, we collect the largest known(to date) dataset of microbiologically confirmed COVID-19 cough sounds from 3,621 individuals. When used in a triaging step within an overall testing protocol, by enabling risk-stratification of individuals before confirmatory tests, our tool can increase the testing capacity of a healthcare system by 43% at disease prevalence of 5%, without additional supplies, trained personnel, or physical infrastructure

Citations (110)

Summary

Insights into COVID-19 Detection through Cough Sounds

The paper, "Cough against COVID: Evidence of COVID-19 Signature in Cough Sounds," explores the potential of utilizing artificial intelligence to detect COVID-19 from cough sounds. This investigation is motivated by the global backlog in conventional COVID-19 testing methodologies, primarily RT-PCR tests, which are resource-intensive and demand substantial logistical efforts. Such challenges are exacerbated in rural and low-resource settings, where access to testing facilities remains limited.

At the core of this research is the hypothesis that cough sounds carry a distinguishable signal capable of indicating COVID-19 infection. The team behind this paper has assembled what is noted to be the largest dataset of COVID-19 cough sounds from 3,621 individuals, with robust statistical evidence achieved by a model delivering an AUC of 0.72 and statistically significant results (p < 0.01). The model's applicability spans to asymptomatic cases, a critical aspect given the surreptitious nature of COVID-19 transmission from individuals who do not exhibit symptoms.

This work employs a convolutional neural network (CNN) architecture to process cough sound spectrograms, with inputs preprocessed using log-mel spectrogram transformations for optimal feature capture. The training regime includes pretraining on non-COVID cough datasets to enhance initial model performance, followed by finetuning with collected COVID-19 data. Such a strategy is essential, leveraging the generalization capabilities of CNNs while mitigating the usual requirement for extensive labeled datasets in audio classification tasks.

An important consideration within the modeling process is the handling of RT-PCR test inaccuracies. The incorporation of label smoothing aids in accounting for potential mislabeling due to test sensitivity shortcomings, thereby yielding a more calibrated model. The processing pipeline also utilizes extensive data augmentation and a carefully curated validation strategy to balance facility-specific biases.

The practical utility of this research is evident in its proposition of a triaging tool. The tool could significantly augment existing healthcare systems' capabilities by theoretically increasing testing throughput by 43% under a disease prevalence rate of 5%. By screening individuals through a non-invasive cough analysis, healthcare systems can prioritize RT-PCR tests for those flagged as high-risk, thereby optimizing resource allocation and potentially curbing virus spread through timely isolation of cases.

In summary, while this paper does not yet claim breakthroughs in diagnostic accuracy of AI-based systems compared to gold-standard testing, it provides valuable insights into scalable, resource-efficient strategies that could be instrumental during pandemics. Future efforts could focus on expanding this dataset, incorporating multimodal data inputs such as breathing sounds, and perhaps refining models for deployment on consumer-grade devices, enabling widespread access to self-screening tools. Such developments could represent a pivotal shift in public health strategies, reducing reliance on traditional testing practices and allowing for more agile and responsive disease monitoring systems.

Youtube Logo Streamline Icon: https://streamlinehq.com