Cross-dataset COVID-19 Transfer Learning with Cough Detection Cough Segmentation and Data Augmentation

2025-04-26 0 0 1.28MB 41 页 10玖币

侵权投诉

Highlights

Cross-dataset COVID-19 Transfer Learning with Cough Detection,

Cough Segmentation, and Data Augmentation

Bagus Tris Atmajaa,b, Zanjabilaa, Suyantoa, Akira Sasoub

•Three processing blocks were proposed to improve cough-based COVID-

19 detection.

•The study reports ablation studies on these three blocks and hyperpa-

rameters tuning.

•It also summarizes previous studies on the same test set of COVID-19

detection.

arXiv:2210.05843v1 [eess.AS] 12 Oct 2022

Cross-dataset COVID-19 Transfer Learning with Cough

Detection, Cough Segmentation, and Data

Augmentation

Bagus Tris Atmajaa,b, Zanjabilaa, Suyantoa, Akira Sasoub

Abstract

This paper addresses issues on cough-based COVID-19 detection. We pro-

pose a cross-dataset transfer learning approach to improve the performance of

COVID-19 detection by incorporating cough detection, cough segmentation,

and data augmentation. The ﬁrst aimed at removing non-cough signals and

cough signals with low probability. The second aimed at segregating several

coughs in a waveform into individual coughs. The third aimed at increasing

the number of samples for the deep learning model. These three processing

blocks are important as our ﬁnding revealed a large margin of improvement

relative to the baseline methods without these blocks. An ablation study is

conducted to optimize hyperparameters and it was found that alpha mixup

is an important factor among others in improving the model performance via

this augmentation method. A summary of this study with previous studies

on the same evaluation set was given to gain insights into diﬀerent methods

of cough-based COVID-19 detection.

Keywords: cough detection, cough segmentation, transfer learning, data

augmentation, COVID-19

Preprint submitted to Computer in Medicine and Biology October 13, 2022

1. Introduction

The coronavirus disease that spread at the end of 2019 in China (COVID-

19) and in early 2020 over the world is showing the unpreparedness of humans

for the pandemic. Although a similar case has occurred previously (SARS,

MERS), the response to diagnose the virus in a short time to prevent its

spread is not optimal. The gold standard polymerase chain reaction (PCR)

test takes time in days and hours. The need for preliminary screening by

using other tools is crucial to avoid the spread of the virus.

The human voice is new blood. For many diseases, a blood test is the

main tool to assess the general condition of the human body. By performing a

blood test, the infection of an organ could be checked; the function of certain

organs could be monitored. Modern medical imaging techniques (fMRI, CT-

scan, X-ray) help medical doctors with accurate diagnoses. On the other

side, the use of humans voice for diagnosing particular diseases is limited.

Nevertheless, acoustic signals from the human body represent the state of

the parts to produce these signals. The sound of the human voice is new

blood that could be used for diagnosing particular diseases like in blood and

imaging.

It has been evidenced that the human voice could be used as the main

tool for diagnosing diseases related to voice production: pathological voice

detection [1], pertussis [2], asthma [3], and respiratory diseases [4]. Moreover,

the use of acoustic analysis has been proven eﬀective for Alzheimer’s disease

classiﬁcation (accuracy of 93.30%) [5] and lung disease (accuracy of 98.92%)

[6]. These examples show the potency of the human voice for diagnosing

diseases, particularly voice-related diseases.

Attempts to explore acoustic analysis beyond voice-related diseases have

been conducted for COVID-19 detection. Several modalities have been tried

including cough, breathing sounds, and speech or voice (including vowels).

The ﬁrst two modalities contain indicators for the symptoms of COVID-19:

continuous cough and shortness of breath. We choose the ﬁrst modality due

to the large available cough data. The cough sounds also showed the highest

speciﬁcity among other modalities in the previous study [7].

Nowadays, computers are part of human daily equipment, ranging from

personal computers to smartphones, which could be used for such acoustic

analysis. Given the beneﬁt of diagnosing particular diseases (COVID-19,

dementia, depression), the development of an application on a smartphone

to detect ﬂare-ups of pulmonary disease [8]. In the future, we predict that

smartphones could be furnished with acoustic analysis applications to detect

diseases like COVID-19 (Figure 1). On that day, we could simply ask our

smartphone if we have COVID-19 or not for preliminary screening, as in

Figure 1. This application will likely be available not only for COVID-19 but

also for other diseases and disorders [8]. In the current stage, applications for

classifying the crying of a baby, counting coughs, and detecting anxiety and

depression are available in the market. All of these applications are based on

acoustic analysis.

To speed up research on cough-based COVID-19 detection, we contribute

three important processing blocks for improving the generalization capability

of COVID-19 detection via transfer learning and cross-dataset evaluation.

These blocks are cough detection, cough segmentation, and data augmen-

tation. In addition to the main contribution above, we also performed an

Alexa, do I

have COVID?

Let me check.

Please cough

on me!

You don’t seems

to have COVID.

Figure 1: Illustration of the smartphone application for COVID-19 detection.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

HighlightsCross-datasetCOVID-19TransferLearningwithCoughDetection,CoughSegmentation,andDataAugmentationBagusTrisAtmajaa;b,Zanjabilaa,Suyantoa,AkiraSasoubThreeprocessingblockswereproposedtoimprovecough-basedCOVID-19detection.Thestudyreportsablationstudiesonthesethreeblocksandhyperpa-rameterstuning....

展开>> 收起<<

Cross-dataset COVID-19 Transfer Learning with Cough Detection Cough Segmentation and Data Augmentation.pdf

共41页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Cross-dataset COVID-19 Transfer Learning with Cough Detection Cough Segmentation and Data Augmentation

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: