Eight pruning deep learning models for low storage and high-speed COVID-19 computed tomography lung segmentation and heatmap-based lesion localization: A multicenter study using COVLIAS 2.0.
AI
COVID-19
Deep learning
Glass ground opacities
Hounsfield units
Lung CT
Lung segmentation
Pruning
Journal
Computers in biology and medicine
ISSN: 1879-0534
Titre abrégé: Comput Biol Med
Pays: United States
ID NLM: 1250250
Informations de publication
Date de publication:
07 2022
07 2022
Historique:
received:
26
02
2022
revised:
05
04
2022
accepted:
26
04
2022
entrez:
25
6
2022
pubmed:
26
6
2022
medline:
29
6
2022
Statut:
ppublish
Résumé
COVLIAS 1.0: an automated lung segmentation was designed for COVID-19 diagnosis. It has issues related to storage space and speed. This study shows that COVLIAS 2.0 uses pruned AI (PAI) networks for improving both storage and speed, wiliest high performance on lung segmentation and lesion localization. ology: The proposed study uses multicenter ∼9,000 CT slices from two different nations, namely, CroMed from Croatia (80 patients, experimental data), and NovMed from Italy (72 patients, validation data). We hypothesize that by using pruning and evolutionary optimization algorithms, the size of the AI models can be reduced significantly, ensuring optimal performance. Eight different pruning techniques (i) differential evolution (DE), (ii) genetic algorithm (GA), (iii) particle swarm optimization algorithm (PSO), and (iv) whale optimization algorithm (WO) in two deep learning frameworks (i) Fully connected network (FCN) and (ii) SegNet were designed. COVLIAS 2.0 was validated using "Unseen NovMed" and benchmarked against MedSeg. Statistical tests for stability and reliability were also conducted. Pruning algorithms (i) FCN-DE, (ii) FCN-GA, (iii) FCN-PSO, and (iv) FCN-WO showed improvement in storage by 92.4%, 95.3%, 98.7%, and 99.8% respectively when compared against solo FCN, and (v) SegNet-DE, (vi) SegNet-GA, (vii) SegNet-PSO, and (viii) SegNet-WO showed improvement by 97.1%, 97.9%, 98.8%, and 99.2% respectively when compared against solo SegNet. AUC > 0.94 (p < 0.0001) on CroMed and > 0.86 (p < 0.0001) on NovMed data set for all eight EA model. PAI <0.25 s per image. DenseNet-121-based Grad-CAM heatmaps showed validation on glass ground opacity lesions. Eight PAI networks that were successfully validated are five times faster, storage efficient, and could be used in clinical settings.
Sections du résumé
BACKGROUND
COVLIAS 1.0: an automated lung segmentation was designed for COVID-19 diagnosis. It has issues related to storage space and speed. This study shows that COVLIAS 2.0 uses pruned AI (PAI) networks for improving both storage and speed, wiliest high performance on lung segmentation and lesion localization.
METHOD
ology: The proposed study uses multicenter ∼9,000 CT slices from two different nations, namely, CroMed from Croatia (80 patients, experimental data), and NovMed from Italy (72 patients, validation data). We hypothesize that by using pruning and evolutionary optimization algorithms, the size of the AI models can be reduced significantly, ensuring optimal performance. Eight different pruning techniques (i) differential evolution (DE), (ii) genetic algorithm (GA), (iii) particle swarm optimization algorithm (PSO), and (iv) whale optimization algorithm (WO) in two deep learning frameworks (i) Fully connected network (FCN) and (ii) SegNet were designed. COVLIAS 2.0 was validated using "Unseen NovMed" and benchmarked against MedSeg. Statistical tests for stability and reliability were also conducted.
RESULTS
Pruning algorithms (i) FCN-DE, (ii) FCN-GA, (iii) FCN-PSO, and (iv) FCN-WO showed improvement in storage by 92.4%, 95.3%, 98.7%, and 99.8% respectively when compared against solo FCN, and (v) SegNet-DE, (vi) SegNet-GA, (vii) SegNet-PSO, and (viii) SegNet-WO showed improvement by 97.1%, 97.9%, 98.8%, and 99.2% respectively when compared against solo SegNet. AUC > 0.94 (p < 0.0001) on CroMed and > 0.86 (p < 0.0001) on NovMed data set for all eight EA model. PAI <0.25 s per image. DenseNet-121-based Grad-CAM heatmaps showed validation on glass ground opacity lesions.
CONCLUSIONS
Eight PAI networks that were successfully validated are five times faster, storage efficient, and could be used in clinical settings.
Identifiants
pubmed: 35751196
pii: S0010-4825(22)00363-8
doi: 10.1016/j.compbiomed.2022.105571
pmc: PMC9123805
pii:
doi:
Types de publication
Journal Article
Multicenter Study
Langues
eng
Sous-ensembles de citation
IM
Pagination
105571Informations de copyright
Copyright © 2022 Elsevier Ltd. All rights reserved.
Références
Eur Rev Med Pharmacol Sci. 2021 Sep;25(18):5876-5884
pubmed: 34604981
Comput Biol Med. 2022 Mar;142:105204
pubmed: 35033879
Clin Chem. 2002 May;48(5):799-801; author reply 801-2
pubmed: 11978620
Insights Imaging. 2011 Apr;2(2):117-132
pubmed: 22347941
IEEE Trans Cybern. 2020 Sep;50(9):3840-3854
pubmed: 32324588
Comput Methods Programs Biomed. 2018 Mar;155:165-177
pubmed: 29512496
Med Biol Eng Comput. 2021 Mar;59(3):511-533
pubmed: 33547549
J Med Syst. 2018 Apr 10;42(5):92
pubmed: 29637403
BMC Med Imaging. 2021 Feb 9;21(1):19
pubmed: 33557772
Int J Cardiovasc Imaging. 2021 Apr;37(4):1349-1360
pubmed: 33211242
Int J Comput Assist Radiol Surg. 2021 Jan;16(1):115-123
pubmed: 33191476
Eur Rev Med Pharmacol Sci. 2021 Aug;25(15):5063-5069
pubmed: 34355379
Med Biol Eng Comput. 2019 Feb;57(2):543-564
pubmed: 30255236
IEEE J Biomed Health Inform. 2021 Nov;25(11):4128-4139
pubmed: 34379599
Diagnostics (Basel). 2021 Dec 02;11(12):
pubmed: 34943494
Eur Radiol. 2020 Oct;30(10):5455-5462
pubmed: 32367422
J Med Virol. 2020 Oct;92(10):1755-1756
pubmed: 32270882
Med Image Anal. 2017 Dec;42:1-13
pubmed: 28732268
Diagnostics (Basel). 2021 Nov 01;11(11):
pubmed: 34829372
Comput Biol Med. 2017 Dec 1;91:198-212
pubmed: 29100114
Eur Rev Med Pharmacol Sci. 2020 Dec;24(23):12609-12622
pubmed: 33336781
Comput Biol Med. 2019 May;108:85-92
pubmed: 31003183
IEEE Trans Med Imaging. 2020 Jun;39(6):1856-1867
pubmed: 31841402
Conf Proc IEEE Eng Med Biol Soc. 2006;2006:6121-4
pubmed: 17945937
Annu Int Conf IEEE Eng Med Biol Soc. 2011;2011:4489-92
pubmed: 22255336
J Med Virol. 2020 Nov;92(11):2312-2313
pubmed: 32383182
Comput Methods Programs Biomed. 2017 Oct;150:9-22
pubmed: 28859832
Cardiovasc Diagn Ther. 2019 Oct;9(5):439-461
pubmed: 31737516
Annu Int Conf IEEE Eng Med Biol Soc. 2012;2012:4446-9
pubmed: 23366914
J Hist Biol. 1975 Fall;8(2):219-41
pubmed: 11609894
Eur J Clin Microbiol Infect Dis. 2020 Jul;39(7):1379-1389
pubmed: 32337662
IEEE Trans Med Imaging. 2020 Nov;39(11):3679-3690
pubmed: 32746113
J Med Syst. 2015 Mar;39(3):22
pubmed: 25666926
J Med Syst. 2017 Aug 23;41(10):152
pubmed: 28836045
Artif Intell Rev. 2022;55(6):4755-4808
pubmed: 35068651
Comput Methods Programs Biomed. 2019 Jul;176:173-193
pubmed: 31200905
Diagnostics (Basel). 2021 Dec 15;11(12):
pubmed: 34943603
Sensors (Basel). 2020 Sep 30;20(19):
pubmed: 33008132
Ultraschall Med. 2014 Jun;35(3):237-45
pubmed: 23258769
Comput Biol Med. 2015 Oct 1;65:54-68
pubmed: 26298486
Ann Med. 2021 Dec;53(1):169-180
pubmed: 33426973
Ann Transl Med. 2021 Jul;9(14):1206
pubmed: 34430647
Proc Inst Mech Eng H. 2013 Jul;227(7):788-98
pubmed: 23636761
Comput Biol Med. 2021 Sep;136:104721
pubmed: 34371320
Eur Rev Med Pharmacol Sci. 2021 Dec;25(24):7997-8003
pubmed: 34982463
World J Diabetes. 2021 Mar 15;12(3):215-237
pubmed: 33758644
Diagnostics (Basel). 2021 Aug 04;11(8):
pubmed: 34441340
Diagnostics (Basel). 2022 Jan 11;12(1):
pubmed: 35054333
Technol Cancer Res Treat. 2013 Dec;12(6):545-57
pubmed: 23745787
Ann Transl Med. 2020 Oct;8(19):1261
pubmed: 33178793
Diagnostics (Basel). 2021 Nov 15;11(11):
pubmed: 34829456
Eur Rev Med Pharmacol Sci. 2021 Oct;25(20):6439-6442
pubmed: 34730226
Comput Biol Med. 2020 Sep;124:103960
pubmed: 32919186
Comput Methods Programs Biomed. 2017 Dec;152:23-34
pubmed: 29054258
Eur J Radiol. 2022 Apr;149:110188
pubmed: 35180580
Acta Biomed. 2020 Mar 19;91(1):157-160
pubmed: 32191675
J Med Syst. 2012 Jun;36(3):1861-71
pubmed: 21243411
PLoS One. 2021 Jun 10;16(6):e0252754
pubmed: 34111168
IEEE Trans Inf Technol Biomed. 2002 Mar;6(1):8-28
pubmed: 11936600
J Neuroradiol. 2018 Mar;45(2):136-141
pubmed: 28987274
Sensors (Basel). 2021 Nov 30;21(23):
pubmed: 34884004
Comput Methods Programs Biomed. 2013 Apr;110(1):66-75
pubmed: 23122720
Annu Int Conf IEEE Eng Med Biol Soc. 2007;2007:335-8
pubmed: 18001958
Eur J Radiol. 2019 May;114:14-24
pubmed: 31005165
Front Biosci (Landmark Ed). 2019 Jan 1;24(3):392-426
pubmed: 30468663
Comput Biol Med. 2021 Mar;130:104210
pubmed: 33550068
Comput Biol Med. 2021 Oct;137:104803
pubmed: 34536856
Front Biosci (Landmark Ed). 2021 Nov 30;26(11):1312-1339
pubmed: 34856770
Comput Math Methods Med. 2022 Feb 1;2022:7672196
pubmed: 35116074
Biochem Med (Zagreb). 2015 Jun 05;25(2):141-51
pubmed: 26110027
Comput Biol Med. 2022 Feb;141:105131
pubmed: 34922173
AJR Am J Roentgenol. 2020 Jul;215(1):87-93
pubmed: 32174129
Radiology. 2020 Aug;296(2):E115-E117
pubmed: 32073353
Int Angiol. 2022 Feb;41(1):9-23
pubmed: 34825801
J Med Syst. 2016 Jun;40(6):142
pubmed: 27114353
J Public Health Res. 2021 Apr 19;10(3):
pubmed: 33876627
IEEE Trans Pattern Anal Mach Intell. 2017 Dec;39(12):2481-2495
pubmed: 28060704
IEEE Trans Neural Netw Learn Syst. 2021 Mar;32(3):932-946
pubmed: 33544680
Acad Radiol. 2020 Dec;27(12):1665-1678
pubmed: 33046370
Eur Rev Med Pharmacol Sci. 2022 Jan;26(1):270-277
pubmed: 35049004
Comput Biol Med. 2017 Dec 1;91:306-317
pubmed: 29107894
Eur Radiol. 2020 Sep;30(9):4893-4902
pubmed: 32300971
Radiology. 2020 Aug;296(2):E41-E45
pubmed: 32049601
Med Biol Eng Comput. 2013 May;51(5):513-23
pubmed: 23292291
J Glob Health. 2018 Jun;8(1):010417
pubmed: 29740501