Factors determining generalization in deep learning models for scoring COVID-CT images

Horry, MJ; Chakraborty, S; Pradhan, B; Fallahpoor, M; Chegeni, H; Paul, M

Factors determining generalization in deep learning models for scoring COVID-CT images

Horry, MJ Chakraborty, S

Pradhan, B Fallahpoor, M Chegeni, H Paul, M

Permalink

Publisher:: American Institute of Mathematical Sciences (AIMS)
Publication Type:: Journal Article
Citation:: Mathematical Biosciences and Engineering, 2021, 18, (6), pp. 9264-9293
Issue Date:: 2021

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download Published versionAdobe PDF (1.97 MB)

View on publisher's site

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Horry, MJ
dc.contributor.author	Chakraborty, S https://orcid.org/0000-0002-0102-5424
dc.contributor.author	Pradhan, B
dc.contributor.author	Fallahpoor, M
dc.contributor.author	Chegeni, H
dc.contributor.author	Paul, M
dc.date.accessioned	2021-10-28T03:52:51Z
dc.date.available	2021-10-28T03:52:51Z
dc.date.issued	2021
dc.identifier.citation	Mathematical Biosciences and Engineering, 2021, 18, (6), pp. 9264-9293
dc.identifier.issn	1551-0018
dc.identifier.uri	http://hdl.handle.net/10453/151249
dc.description.abstract	<jats:p xml:lang="fr"><abstract> <p>The COVID-19 pandemic has inspired unprecedented data collection and computer vision modelling efforts worldwide, focused on the diagnosis of COVID-19 from medical images. However, these models have found limited, if any, clinical application due in part to unproven generalization to data sets beyond their source training corpus. This study investigates the generalizability of deep learning models using publicly available COVID-19 Computed Tomography data through cross dataset validation. The predictive ability of these models for COVID-19 severity is assessed using an independent dataset that is stratified for COVID-19 lung involvement. Each inter-dataset study is performed using histogram equalization, and contrast limited adaptive histogram equalization with and without a learning Gabor filter. We show that under certain conditions, deep learning models can generalize well to an external dataset with F1 scores up to 86%. The best performing model shows predictive accuracy of between 75% and 96% for lung involvement scoring against an external expertly stratified dataset. From these results we identify key factors promoting deep learning generalization, being primarily the uniform acquisition of training images, and secondly diversity in CT slice position.</p> </abstract></jats:p>
dc.language	en
dc.publisher	American Institute of Mathematical Sciences (AIMS)
dc.relation.ispartof	Mathematical Biosciences and Engineering
dc.relation.isbasedon	10.3934/mbe.2021456
dc.rights	info:eu-repo/semantics/openAccess
dc.subject	0102 Applied Mathematics, 0903 Biomedical Engineering, 0904 Chemical Engineering
dc.subject.classification	Bioinformatics
dc.title	Factors determining generalization in deep learning models for scoring COVID-CT images
dc.type	Journal Article
utslib.citation.volume	18
utslib.for	0102 Applied Mathematics
utslib.for	0903 Biomedical Engineering
utslib.for	0904 Chemical Engineering
pubs.organisational-group	/University of Technology Sydney
pubs.organisational-group	/University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	/University of Technology Sydney/Faculty of Engineering and Information Technology/School of Information, Systems and Modelling
utslib.copyright.status	open_access	*
dc.date.updated	2021-10-28T03:52:49Z
pubs.issue	6
pubs.publication-status	Published
pubs.volume	18
utslib.citation.issue	6

Abstract:

<abstract> <p>The COVID-19 pandemic has inspired unprecedented data collection and computer vision modelling efforts worldwide, focused on the diagnosis of COVID-19 from medical images. However, these models have found limited, if any, clinical application due in part to unproven generalization to data sets beyond their source training corpus. This study investigates the generalizability of deep learning models using publicly available COVID-19 Computed Tomography data through cross dataset validation. The predictive ability of these models for COVID-19 severity is assessed using an independent dataset that is stratified for COVID-19 lung involvement. Each inter-dataset study is performed using histogram equalization, and contrast limited adaptive histogram equalization with and without a learning Gabor filter. We show that under certain conditions, deep learning models can generalize well to an external dataset with F1 scores up to 86%. The best performing model shows predictive accuracy of between 75% and 96% for lung involvement scoring against an external expertly stratified dataset. From these results we identify key factors promoting deep learning generalization, being primarily the uniform acquisition of training images, and secondly diversity in CT slice position.</p> </abstract>

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/151249