Fully automated volumetric measurement of malignant pleural mesothelioma by deep learning AI: validation and comparison with modified RECIST response criteria.
imaging/CT MRI
mesothelioma
pleural disease
Journal
Thorax
ISSN: 1468-3296
Titre abrégé: Thorax
Pays: England
ID NLM: 0417353
Informations de publication
Date de publication:
12 2022
12 2022
Historique:
received:
15
06
2021
accepted:
20
12
2021
pubmed:
4
2
2022
medline:
19
11
2022
entrez:
3
2
2022
Statut:
ppublish
Résumé
In malignant pleural mesothelioma (MPM), complex tumour morphology results in inconsistent radiological response assessment. Promising volumetric methods require automation to be practical. We developed a fully automated Convolutional Neural Network (CNN) for this purpose, performed blinded validation and compared CNN and human response classification and survival prediction in patients treated with chemotherapy. In a multicentre retrospective cohort study; 183 CT datasets were split into training and internal validation (123 datasets (80 fully annotated); 108 patients; 1 centre) and external validation (60 datasets (all fully annotated); 30 patients; 3 centres). Detailed manual annotations were used to train the CNN, which used two-dimensional U-Net architecture. CNN performance was evaluated using correlation, Bland-Altman and Dice agreement. Volumetric response/progression were defined as ≤30%/≥20% change and compared with modified Response Evaluation Criteria In Solid Tumours (mRECIST) by Cohen's kappa. Survival was assessed using Kaplan-Meier methodology. Human and artificial intelligence (AI) volumes were strongly correlated (validation set r=0.851, p<0.0001). Agreement was strong (validation set mean bias +31 cm We have developed and validated the first fully automated CNN for volumetric MPM segmentation. CNN performance may be further improved by enriching future training sets with morphologically challenging features. Volumetric response thresholds require further calibration in future studies.
Sections du résumé
BACKGROUND
In malignant pleural mesothelioma (MPM), complex tumour morphology results in inconsistent radiological response assessment. Promising volumetric methods require automation to be practical. We developed a fully automated Convolutional Neural Network (CNN) for this purpose, performed blinded validation and compared CNN and human response classification and survival prediction in patients treated with chemotherapy.
METHODS
In a multicentre retrospective cohort study; 183 CT datasets were split into training and internal validation (123 datasets (80 fully annotated); 108 patients; 1 centre) and external validation (60 datasets (all fully annotated); 30 patients; 3 centres). Detailed manual annotations were used to train the CNN, which used two-dimensional U-Net architecture. CNN performance was evaluated using correlation, Bland-Altman and Dice agreement. Volumetric response/progression were defined as ≤30%/≥20% change and compared with modified Response Evaluation Criteria In Solid Tumours (mRECIST) by Cohen's kappa. Survival was assessed using Kaplan-Meier methodology.
RESULTS
Human and artificial intelligence (AI) volumes were strongly correlated (validation set r=0.851, p<0.0001). Agreement was strong (validation set mean bias +31 cm
CONCLUSION
We have developed and validated the first fully automated CNN for volumetric MPM segmentation. CNN performance may be further improved by enriching future training sets with morphologically challenging features. Volumetric response thresholds require further calibration in future studies.
Identifiants
pubmed: 35110367
pii: thoraxjnl-2021-217808
doi: 10.1136/thoraxjnl-2021-217808
pmc: PMC9685726
doi:
Types de publication
Journal Article
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
1251-1259Subventions
Organisme : Chief Scientist Office
ID : ETM/285
Pays : United Kingdom
Informations de copyright
© Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.
Déclaration de conflit d'intérêts
Competing interests: ACK, GWC, ME, ST and KGB have no conflicts of interest to declare. OA, AJW, JV and KAG are employees of Canon Medical Research Europe.
Références
Lancet. 2021 Jan 30;397(10272):375-386
pubmed: 33485464
Lancet. 2016 Apr 2;387(10026):1405-1414
pubmed: 26719230
J Thorac Cardiovasc Surg. 1998 Feb;115(2):310-7; discussion 317-8
pubmed: 9475525
Transl Oncol. 2012 Feb;5(1):19-25
pubmed: 22348172
J Clin Oncol. 2003 Jul 15;21(14):2636-44
pubmed: 12860938
ANZ J Surg. 2012 Oct;82(10):737-41
pubmed: 22957887
Lung Cancer. 2020 Dec;150:12-20
pubmed: 33039775
Eur J Cancer. 2012 Nov;48(16):2983-92
pubmed: 22704893
Eur J Cardiothorac Surg. 2003 Dec;24(6):1019-24
pubmed: 14643823
Int J Comput Assist Radiol Surg. 2017 Apr;12(4):529-538
pubmed: 28028655
Trials. 2018 Apr 18;19(1):233
pubmed: 29669604
Eur Respir J. 2011 Jul;38(1):162-8
pubmed: 21273389
J Thorac Oncol. 2008 Jul;3(7):756-63
pubmed: 18594322
Med Phys. 2004 May;31(5):1105-15
pubmed: 15191298
BMJ Qual Saf. 2016 Dec;25(12):986-992
pubmed: 26369893
Br J Clin Pharmacol. 2012 Jun;73(6):846-54
pubmed: 22242836
Comput Med Imaging Graph. 2018 Apr;65:79-92
pubmed: 28789867
Cancer Imaging. 2011 Oct 31;11:155-62
pubmed: 22042236
Lung Cancer. 2006 May;52(2):141-8
pubmed: 16530882
J Med Imaging (Bellingham). 2018 Jul;5(3):034503
pubmed: 30840717
J Thorac Oncol. 2014 Aug;9(8):1187-94
pubmed: 25157772
Ann Thorac Surg. 2016 Oct;102(4):1059-66
pubmed: 27596916
J Med Imaging (Bellingham). 2020 Jan;7(1):012705
pubmed: 32016133
J Thorac Oncol. 2021 Oct;16(10):1705-1717
pubmed: 34116230
Ann Oncol. 2004 Feb;15(2):257-60
pubmed: 14760119
Med Phys. 2011 Jan;38(1):238-44
pubmed: 21361192
Invest Radiol. 2008 Oct;43(10):737-44
pubmed: 18791416
Eur J Radiol. 2010 Apr;74(1):130-5
pubmed: 19268516
J Thorac Oncol. 2016 Aug;11(8):1335-1344
pubmed: 27180318