Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Exploring Automatic Diagnosis of COVID-19 from Crowdsourced Respiratory Sound Data (2006.05919v3)

Published 10 Jun 2020 in cs.SD, cs.LG, and eess.AS

Abstract: Audio signals generated by the human body (e.g., sighs, breathing, heart, digestion, vibration sounds) have routinely been used by clinicians as indicators to diagnose disease or assess disease progression. Until recently, such signals were usually collected through manual auscultation at scheduled visits. Research has now started to use digital technology to gather bodily sounds (e.g., from digital stethoscopes) for cardiovascular or respiratory examination, which could then be used for automatic analysis. Some initial work shows promise in detecting diagnostic signals of COVID-19 from voice and coughs. In this paper we describe our data analysis over a large-scale crowdsourced dataset of respiratory sounds collected to aid diagnosis of COVID-19. We use coughs and breathing to understand how discernible COVID-19 sounds are from those in asthma or healthy controls. Our results show that even a simple binary machine learning classifier is able to classify correctly healthy and COVID-19 sounds. We also show how we distinguish a user who tested positive for COVID-19 and has a cough from a healthy user with a cough, and users who tested positive for COVID-19 and have a cough from users with asthma and a cough. Our models achieve an AUC of above 80% across all tasks. These results are preliminary and only scratch the surface of the potential of this type of data and audio-based machine learning. This work opens the door to further investigation of how automatically analysed respiratory patterns could be used as pre-screening signals to aid COVID-19 diagnosis.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (9)
  1. Chloë Brown (11 papers)
  2. Jagmohan Chauhan (23 papers)
  3. Andreas Grammenos (16 papers)
  4. Jing Han (60 papers)
  5. Apinan Hasthanasombat (6 papers)
  6. Dimitris Spathis (35 papers)
  7. Tong Xia (25 papers)
  8. Pietro Cicuta (28 papers)
  9. Cecilia Mascolo (86 papers)
Citations (401)

Summary

Essay on "Exploring Automatic Diagnosis of COVID-19 from Crowdsourced Respiratory Sound Data"

The academic paper titled "Exploring Automatic Diagnosis of COVID-19 from Crowdsourced Respiratory Sound Data" presents a detailed examination of using human respiratory sounds as diagnostic markers for COVID-19. This investigation leverages a substantial dataset obtained through crowdsourcing, specifically targeting the collection of audio signals such as coughs and breathing from individuals across diverse geographical locations. The paper is significant within the landscape of digital health diagnostics, offering insights into the utilization of machine learning to distinguish COVID-19 respiratory acoustic signatures from those of healthy individuals or individuals with other conditions such as asthma.

The research involves a methodical approach to data collection and analysis. Using a web-based and Android app, the authors gathered over 10,000 samples from approximately 7,000 unique users around the globe. This dataset includes digital audio samples that were manually pre-processed to ensure quality, and subsequent feature extraction was performed both through handcrafted methods and transfer learning techniques, notably utilizing the VGGish model.

A salient aspect of this research is the focus on binary classification tasks, attempting to distinguish between healthy individuals and those who have tested positive for COVID-19, as well as differentiating COVID-19 positive individuals with a cough from those with a cough from asthma. The paper employs several models, primarily focusing on logistic regression and support vector machines, to achieve classification tasks with an Area Under Curve (AUC) performance exceeding 80% for all tasks. The data processing pipeline incorporated feature dimensionality reduction via PCA, which served to optimize model efficacy further.

While the research achieved promising initial results, these findings are contextualized within the limitations imposed by the self-reported nature of the crowdsourced data, the uncontrolled environmental factors affecting sound recordings, and the lack of clinical ground truth for the COVID-19 status in many instances. However, the findings clearly demonstrate that respiratory sound patterns hold potential as diagnostic signals, warranting further exploration.

Practically, this paper contributes to the ongoing efforts to incorporate non-invasive, wide-scale screening tools into public health strategy, particularly in light of the constraints faced during the COVID-19 pandemic. The capacity to remotely gather and automatically analyze health data over ubiquitous smart devices aligns with the digital transformation in healthcare, potentially offering a scalable and accessible pre-screening mechanism.

Theoretically, the paper reinforces the concept that bodily sounds can act as viable biomarkers for disease detection, opening pathways for multifaceted investigations into the diagnostic capability of audio-signals. This consideration extends beyond COVID-19, positing the broader application of audio-based diagnosis in detecting other diseases with respiratory or vocal manifestations.

Looking toward the future, further research is suggested to refine and validate these preliminary results. Incremental data collection could ensure more robust models leveraging deep learning frameworks, provided the dataset expands significantly and integrates clinically validated instances of disease status. Additionally, incorporating longitudinal data from returning app users could facilitate studies on disease progression, potentially extending the utility of these signals as dynamic biomarkers.

Overall, this paper underscores the potential of automated, sound-based machine learning systems in enhancing health diagnostics and proposes methodologies that could be incrementally developed into highly effective screening tools. This work exemplifies an innovative interconnection between machine learning, acoustic analysis, and telehealth—offering a strategic addition to global health responses amid pandemics and beyond.