Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging
IntroductionAlthough medical imaging plays a crucial role in stroke management, machine learning (ML) has been increasingly used in this field, particularly in lesion segmentation. Despite advances in acquisition technologies and segmentation architectures, one of the main challenges of subacute str...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-02-01
|
| Series: | Frontiers in Neurology |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fneur.2025.1434334/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1823860323370663936 |
|---|---|
| author | Juliette Moreau Juliette Moreau Laura Mechtouff Laura Mechtouff David Rousseau Omer Faruk Eker Omer Faruk Eker Yves Berthezene Yves Berthezene Tae-Hee Cho Tae-Hee Cho Carole Frindel Carole Frindel |
| author_facet | Juliette Moreau Juliette Moreau Laura Mechtouff Laura Mechtouff David Rousseau Omer Faruk Eker Omer Faruk Eker Yves Berthezene Yves Berthezene Tae-Hee Cho Tae-Hee Cho Carole Frindel Carole Frindel |
| author_sort | Juliette Moreau |
| collection | DOAJ |
| description | IntroductionAlthough medical imaging plays a crucial role in stroke management, machine learning (ML) has been increasingly used in this field, particularly in lesion segmentation. Despite advances in acquisition technologies and segmentation architectures, one of the main challenges of subacute stroke lesion segmentation in computed tomography (CT) imaging is image contrast.MethodsTo address this issue, we propose a method to assess the contrast quality of an image dataset with a ML trained model for segmentation. This method identifies the critical contrast level below which the medical-imaging model fails to learn meaningful content from images. Contrast measurement relies on the Fisher's ratio, estimating how well the stroke lesion is contrasted from the background. The critical contrast is found-thanks to the following three methods: Performance, graphical, and clustering analysis. Defining this threshold improves dataset design and accelerates training by excluding low-contrast images.ResultsApplication of this method to brain lesion segmentation in CT imaging highlights a Fisher's ratio threshold value of 0.05, and training validation of a new model without these images confirms this with similar results with only 60% of the training data, resulting in an almost 30% reduction in initial training time. Moreover, the model trained without the low-contrast images performed equally well with all images when tested on another database.DiscussionThis study opens discussion with clinicians concerning the limitations, areas for improvement, and strategies for enhancing datasets and training models. While the methodology was only applied to stroke lesion segmentation in CT images, it has the potential to be adapted to other tasks. |
| format | Article |
| id | doaj-art-059344d89ad24aa9b4c814f7baebc2fb |
| institution | Kabale University |
| issn | 1664-2295 |
| language | English |
| publishDate | 2025-02-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Neurology |
| spelling | doaj-art-059344d89ad24aa9b4c814f7baebc2fb2025-02-10T16:05:17ZengFrontiers Media S.A.Frontiers in Neurology1664-22952025-02-011610.3389/fneur.2025.14343341434334Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imagingJuliette Moreau0Juliette Moreau1Laura Mechtouff2Laura Mechtouff3David Rousseau4Omer Faruk Eker5Omer Faruk Eker6Yves Berthezene7Yves Berthezene8Tae-Hee Cho9Tae-Hee Cho10Carole Frindel11Carole Frindel12CarMeN, INSERM U1060, INRAe U1397, Université Lyon 1, INSA de Lyon, Pierre-Bénite, FranceCREATIS, Universite Claude Bernard Lyon 1, INSA Lyon, UMR CNRS 5220, Inserm U1294, Villeurbanne, FranceCarMeN, INSERM U1060, INRAe U1397, Université Lyon 1, INSA de Lyon, Pierre-Bénite, FranceDepartment of Neurology, Hospices Civils de Lyon, Bron, FranceLARIS, UMR IRHS INRAe, Universite d'Angers, Angers, FranceCREATIS, Universite Claude Bernard Lyon 1, INSA Lyon, UMR CNRS 5220, Inserm U1294, Villeurbanne, FranceDepartment of Neurology, Hospices Civils de Lyon, Bron, FranceCREATIS, Universite Claude Bernard Lyon 1, INSA Lyon, UMR CNRS 5220, Inserm U1294, Villeurbanne, FranceDepartment of Neurology, Hospices Civils de Lyon, Bron, FranceCarMeN, INSERM U1060, INRAe U1397, Université Lyon 1, INSA de Lyon, Pierre-Bénite, FranceDepartment of Neurology, Hospices Civils de Lyon, Bron, FranceCREATIS, Universite Claude Bernard Lyon 1, INSA Lyon, UMR CNRS 5220, Inserm U1294, Villeurbanne, FranceInstitut Universitaire de France (IUF), Paris, FranceIntroductionAlthough medical imaging plays a crucial role in stroke management, machine learning (ML) has been increasingly used in this field, particularly in lesion segmentation. Despite advances in acquisition technologies and segmentation architectures, one of the main challenges of subacute stroke lesion segmentation in computed tomography (CT) imaging is image contrast.MethodsTo address this issue, we propose a method to assess the contrast quality of an image dataset with a ML trained model for segmentation. This method identifies the critical contrast level below which the medical-imaging model fails to learn meaningful content from images. Contrast measurement relies on the Fisher's ratio, estimating how well the stroke lesion is contrasted from the background. The critical contrast is found-thanks to the following three methods: Performance, graphical, and clustering analysis. Defining this threshold improves dataset design and accelerates training by excluding low-contrast images.ResultsApplication of this method to brain lesion segmentation in CT imaging highlights a Fisher's ratio threshold value of 0.05, and training validation of a new model without these images confirms this with similar results with only 60% of the training data, resulting in an almost 30% reduction in initial training time. Moreover, the model trained without the low-contrast images performed equally well with all images when tested on another database.DiscussionThis study opens discussion with clinicians concerning the limitations, areas for improvement, and strategies for enhancing datasets and training models. While the methodology was only applied to stroke lesion segmentation in CT images, it has the potential to be adapted to other tasks.https://www.frontiersin.org/articles/10.3389/fneur.2025.1434334/fulldeep learningsegmentationquality controlcontrast analysisstrokeCT imaging |
| spellingShingle | Juliette Moreau Juliette Moreau Laura Mechtouff Laura Mechtouff David Rousseau Omer Faruk Eker Omer Faruk Eker Yves Berthezene Yves Berthezene Tae-Hee Cho Tae-Hee Cho Carole Frindel Carole Frindel Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging Frontiers in Neurology deep learning segmentation quality control contrast analysis stroke CT imaging |
| title | Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging |
| title_full | Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging |
| title_fullStr | Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging |
| title_full_unstemmed | Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging |
| title_short | Contrast quality control for segmentation task based on deep learning models—Application to stroke lesion in CT imaging |
| title_sort | contrast quality control for segmentation task based on deep learning models application to stroke lesion in ct imaging |
| topic | deep learning segmentation quality control contrast analysis stroke CT imaging |
| url | https://www.frontiersin.org/articles/10.3389/fneur.2025.1434334/full |
| work_keys_str_mv | AT juliettemoreau contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT juliettemoreau contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT lauramechtouff contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT lauramechtouff contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT davidrousseau contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT omerfarukeker contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT omerfarukeker contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT yvesberthezene contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT yvesberthezene contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT taeheecho contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT taeheecho contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT carolefrindel contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging AT carolefrindel contrastqualitycontrolforsegmentationtaskbasedondeeplearningmodelsapplicationtostrokelesioninctimaging |