Development of a deep learning model for classifying thymoma as Masaoka-Koga stage I or II via preoperative CT images.
Masaoka-Koga stage (MK stage)
Thymoma
X-ray computed tomography (X-ray CT)
artificial intelligence (AI) (computer vision systems)
neural networks
Journal
Annals of translational medicine
ISSN: 2305-5839
Titre abrégé: Ann Transl Med
Pays: China
ID NLM: 101617978
Informations de publication
Date de publication:
Mar 2020
Mar 2020
Historique:
entrez:
2
5
2020
pubmed:
2
5
2020
medline:
2
5
2020
Statut:
ppublish
Résumé
Accurate thymoma staging via computed tomography (CT) images is difficult even for experienced thoracic doctors. Here we developed a preoperative staging tool differentiating Masaoka-Koga (MK) stage I patients from stage II patients using CT images. CT images of 174 thymoma patients were retrospectively selected. Two chest radiologists independently assessed the images. Variables with statistical differences in univariate analysis were adjusted for age, sex, and smoking history in multivariate logical regression to determine independent predictors of the thymoma stage. We established a deep learning (DL) 3D-DenseNet model to distinguish the MK stage I and stage II thymomas. Furthermore, we compared two different methods to label the regions of interest (ROI) in CT images. In routine CT images, there were statistical differences (P<0.05) in contour, necrosis, cystic components, and the degree of enhancement between stage I and II disease. Multivariate logical regression showed that only the degree of enhancement was an independent predictor of the thymoma stage. The area under the receiver operating characteristic curve (AUC) of routine CT images for classifying thymoma as MK stage I or II was low (AUC =0.639). The AUC of the 3D-DenseNet model showed better performance with a higher AUC (0.773). ROIs outlined by segmentation labels performed better (AUC =0.773) than those outlined by bounding box labels (AUC =0.722). Our DL 3D-DenseNet may aid thymoma stage classification, which may ultimately guide surgical treatment and improve outcomes. Compared with conventional methods, this approach provides improved staging accuracy. Moreover, ROIs labeled by segmentation is more recommendable when the sample size is limited.
Sections du résumé
BACKGROUND
BACKGROUND
Accurate thymoma staging via computed tomography (CT) images is difficult even for experienced thoracic doctors. Here we developed a preoperative staging tool differentiating Masaoka-Koga (MK) stage I patients from stage II patients using CT images.
METHODS
METHODS
CT images of 174 thymoma patients were retrospectively selected. Two chest radiologists independently assessed the images. Variables with statistical differences in univariate analysis were adjusted for age, sex, and smoking history in multivariate logical regression to determine independent predictors of the thymoma stage. We established a deep learning (DL) 3D-DenseNet model to distinguish the MK stage I and stage II thymomas. Furthermore, we compared two different methods to label the regions of interest (ROI) in CT images.
RESULTS
RESULTS
In routine CT images, there were statistical differences (P<0.05) in contour, necrosis, cystic components, and the degree of enhancement between stage I and II disease. Multivariate logical regression showed that only the degree of enhancement was an independent predictor of the thymoma stage. The area under the receiver operating characteristic curve (AUC) of routine CT images for classifying thymoma as MK stage I or II was low (AUC =0.639). The AUC of the 3D-DenseNet model showed better performance with a higher AUC (0.773). ROIs outlined by segmentation labels performed better (AUC =0.773) than those outlined by bounding box labels (AUC =0.722).
CONCLUSIONS
CONCLUSIONS
Our DL 3D-DenseNet may aid thymoma stage classification, which may ultimately guide surgical treatment and improve outcomes. Compared with conventional methods, this approach provides improved staging accuracy. Moreover, ROIs labeled by segmentation is more recommendable when the sample size is limited.
Identifiants
pubmed: 32355731
doi: 10.21037/atm.2020.02.183
pii: atm-08-06-287
pmc: PMC7186715
doi:
Types de publication
Journal Article
Langues
eng
Pagination
287Informations de copyright
2020 Annals of Translational Medicine. All rights reserved.
Déclaration de conflit d'intérêts
Conflicts of Interest: The authors have no conflicts of interest to declare.
Références
Ann Thorac Surg. 2015 Feb;99(2):462-70
pubmed: 25534526
J Thorac Oncol. 2017 Apr;12(4):734-744
pubmed: 28126540
J Thorac Dis. 2017 Aug;9(8):2369-2374
pubmed: 28932541
Eur J Cardiothorac Surg. 2012 Mar;41(3):623-7
pubmed: 22011779
EClinicalMedicine. 2019 Mar 17;9:52-59
pubmed: 31143882
Cell. 2018 Feb 22;172(5):1122-1131.e9
pubmed: 29474911
Acta Radiol. 2016 Nov;57(11):1318-1325
pubmed: 26089525
Eur J Cardiothorac Surg. 2014 Sep;46(3):361-8
pubmed: 24482389
Nature. 2015 May 28;521(7553):436-44
pubmed: 26017442
J Cardiothorac Surg. 2014 Mar 19;9:51
pubmed: 24646138
Eur J Cardiothorac Surg. 2015 Nov;48(5):698-704
pubmed: 25527169
Ann Thorac Surg. 2016 Feb;101(2):520-6
pubmed: 26482784
J Thorac Oncol. 2015 Apr;10(4):691-700
pubmed: 25629638
Lung Cancer. 2010 Jun;68(3):460-5
pubmed: 19717204
Jpn J Clin Oncol. 2007 Feb;37(2):146-9
pubmed: 17337514
J Thorac Oncol. 2011 Jul;6(7 Suppl 3):S1710-6
pubmed: 21847052
J Thorac Oncol. 2011 Jul;6(7):1274-81
pubmed: 21623235
Eur J Cardiothorac Surg. 2016 Apr;49(4):1144-9
pubmed: 26324679
J Thorac Dis. 2016 Apr;8(4):680-6
pubmed: 27114835
J Thorac Oncol. 2017 Jan;12(1):129-136
pubmed: 27566187
J Thorac Oncol. 2013 Jul;8(7):952-8
pubmed: 23594467
Radiol Med. 2010 Feb;115(1):1-21
pubmed: 20017005
Eur J Cardiothorac Surg. 2014 Jun;45(6):1020-7
pubmed: 24459215
J Thorac Oncol. 2011 Jul;6(7 Suppl 3):S1717-23
pubmed: 21847053
Nat Med. 2019 Jan;25(1):30-36
pubmed: 30617336
Med Image Anal. 2017 Dec;42:60-88
pubmed: 28778026
Genes (Basel). 2018 Jul 30;9(8):
pubmed: 30061525
Eur J Cardiothorac Surg. 2011 Jul;40(1):146-53
pubmed: 21093283
Chest. 2005 Mar;127(3):755-61
pubmed: 15764754
J Thorac Dis. 2016 May;8(5):901-10
pubmed: 27162665
Acad Radiol. 2018 Nov;25(11):1472-1480
pubmed: 29606338
J Thorac Dis. 2016 Apr;8(4):727-37
pubmed: 27114841
J Thorac Oncol. 2010 Oct;5(10 Suppl 4):S318-22
pubmed: 20859126
Surg Endosc. 2008 May;22(5):1272-7
pubmed: 17943356
Eur J Cardiothorac Surg. 2016 Jun;49(6):1545-52
pubmed: 26670806