Unsupervised Deep Learning for Stroke Lesion Segmentation on Follow-up CT Based on Generative Adversarial Networks.
Journal
AJNR. American journal of neuroradiology
ISSN: 1936-959X
Titre abrégé: AJNR Am J Neuroradiol
Pays: United States
ID NLM: 8003708
Informations de publication
Date de publication:
08 2022
08 2022
Historique:
received:
27
01
2022
accepted:
02
06
2022
pubmed:
29
7
2022
medline:
21
3
2023
entrez:
28
7
2022
Statut:
ppublish
Résumé
Supervised deep learning is the state-of-the-art method for stroke lesion segmentation on NCCT. Supervised methods require manual lesion annotations for model development, while unsupervised deep learning methods such as generative adversarial networks do not. The aim of this study was to develop and evaluate a generative adversarial network to segment infarct and hemorrhagic stroke lesions on follow-up NCCT scans. Training data consisted of 820 patients with baseline and follow-up NCCT from 3 Dutch acute ischemic stroke trials. A generative adversarial network was optimized to transform a follow-up scan with a lesion to a generated baseline scan without a lesion by generating a difference map that was subtracted from the follow-up scan. The generated difference map was used to automatically extract lesion segmentations. Segmentation of primary hemorrhagic lesions, hemorrhagic transformation of ischemic stroke, and 24-hour and 1-week follow-up infarct lesions were evaluated relative to expert annotations with the Dice similarity coefficient, Bland-Altman analysis, and intraclass correlation coefficient. The median Dice similarity coefficient was 0.31 (interquartile range, 0.08-0.59) and 0.59 (interquartile range, 0.29-0.74) for the 24-hour and 1-week infarct lesions, respectively. A much lower Dice similarity coefficient was measured for hemorrhagic transformation (median, 0.02; interquartile range, 0-0.14) and primary hemorrhage lesions (median, 0.08; interquartile range, 0.01-0.35). Predicted lesion volume and the intraclass correlation coefficient were good for the 24-hour (bias, 3 mL; limits of agreement, -64-59 mL; intraclass correlation coefficient, 0.83; 95% CI, 0.78-0.88) and excellent for the 1-week (bias, -4 m; limits of agreement,-66-58 mL; intraclass correlation coefficient, 0.90; 95% CI, 0.83-0.93) follow-up infarct lesions. An unsupervised generative adversarial network can be used to obtain automated infarct lesion segmentations with a moderate Dice similarity coefficient and good volumetric correspondence.
Sections du résumé
BACKGROUND AND PURPOSE
Supervised deep learning is the state-of-the-art method for stroke lesion segmentation on NCCT. Supervised methods require manual lesion annotations for model development, while unsupervised deep learning methods such as generative adversarial networks do not. The aim of this study was to develop and evaluate a generative adversarial network to segment infarct and hemorrhagic stroke lesions on follow-up NCCT scans.
MATERIALS AND METHODS
Training data consisted of 820 patients with baseline and follow-up NCCT from 3 Dutch acute ischemic stroke trials. A generative adversarial network was optimized to transform a follow-up scan with a lesion to a generated baseline scan without a lesion by generating a difference map that was subtracted from the follow-up scan. The generated difference map was used to automatically extract lesion segmentations. Segmentation of primary hemorrhagic lesions, hemorrhagic transformation of ischemic stroke, and 24-hour and 1-week follow-up infarct lesions were evaluated relative to expert annotations with the Dice similarity coefficient, Bland-Altman analysis, and intraclass correlation coefficient.
RESULTS
The median Dice similarity coefficient was 0.31 (interquartile range, 0.08-0.59) and 0.59 (interquartile range, 0.29-0.74) for the 24-hour and 1-week infarct lesions, respectively. A much lower Dice similarity coefficient was measured for hemorrhagic transformation (median, 0.02; interquartile range, 0-0.14) and primary hemorrhage lesions (median, 0.08; interquartile range, 0.01-0.35). Predicted lesion volume and the intraclass correlation coefficient were good for the 24-hour (bias, 3 mL; limits of agreement, -64-59 mL; intraclass correlation coefficient, 0.83; 95% CI, 0.78-0.88) and excellent for the 1-week (bias, -4 m; limits of agreement,-66-58 mL; intraclass correlation coefficient, 0.90; 95% CI, 0.83-0.93) follow-up infarct lesions.
CONCLUSIONS
An unsupervised generative adversarial network can be used to obtain automated infarct lesion segmentations with a moderate Dice similarity coefficient and good volumetric correspondence.
Identifiants
pubmed: 35902122
pii: ajnr.A7582
doi: 10.3174/ajnr.A7582
pmc: PMC9575413
doi:
Types de publication
Journal Article
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
1107-1114Investigateurs
Diederik Dippel
(D)
Charles Majoie
(C)
Rick van Nuland
(R)
Charles Majoie
(C)
Aad van der Lugt
(A)
Adriaan van Es
(A)
Pieter-Jan van Doormaal
(PJ)
René van den Berg
(R)
Ludo Beenen
(L)
Bart Emmer
(B)
Stefan Roosendaal
(S)
Wim van Zwam
(W)
Alida Annechien Postma
(AA)
Lonneke Yo
(L)
Menno Krietemeijer
(M)
Geert Lycklama
(G)
Jasper Martens
(J)
Sebastiaan Hammer
(S)
Anton Meijer
(A)
Reinoud Bokkers
(R)
Anouk van der Hoorn
(A)
Ido van den Wijngaard
(I)
Albert Yoo
(A)
Dick Gerrits
(D)
Robert van Oostenbrugge
(R)
Bart Emmer
(B)
Jonathan M Coutinho
(J)
Martine Truijman
(M)
Julie Staals
(J)
Bart van der Worp
(B)
J Boogaarts
(J)
Ben Jansen
(B)
Sanne Zinkstok
(S)
Yvo Roos
(Y)
Peter Koudstaal
(P)
Diederik Dippel
(D)
Jonathan M Coutinho
(J)
Koos Keizer
(K)
Sanne Manschot
(S)
Jelis Boiten
(J)
Henk Kerkhoff
(H)
Ido van den Wijngaard
(I)
Hester Lingsma
(H)
Diederik Dippel
(D)
Vicky Chalos
(V)
Olvert Berkhemer
(O)
Aad van der Lugt
(A)
Charles Majoie
(C)
Adriaan Versteeg
(A)
Lennard Wolff
(L)
Matthijs van der Sluijs
(M)
Henk van Voorst
(H)
Manon Tolhuisen
(M)
Hugo Ten Cate
(H)
Moniek de Maat
(M)
Samantha Donse-Donkel
(S)
Heleen van Beusekom
(H)
Aladdin Taha
(A)
Aarazo Barakzie
(A)
Vicky Chalos
(V)
Rob van de Graaf
(R)
Wouter van der Steen
(W)
Aladdin Taha
(A)
Samantha Donse-Donkel
(S)
Lennard Wolff
(L)
Kilian Treurniet
(K)
Sophie van den Berg
(S)
Natalie LeCouffe
(N)
Manon Kappelhof
(M)
Rik Reinink
(R)
Manon Tolhuisen
(M)
Leon Rinkel
(L)
Josje Brouwer
(J)
Agnetha Bruggeman
(A)
Henk van Voorst
(H)
Robert-Jan Goldhoorn
(RJ)
Wouter Hinsenveld
(W)
Anne Pirson
(A)
Susan Olthuis
(S)
Simone Uniken Venema
(S)
Sjan Teeselink
(S)
Lotte Sondag
(L)
Sabine Collette
(S)
Martin Sterrenberg
(M)
Naziha El Ghannouti
(N)
Laurine van der Steen
(L)
Sabrina Verheesen
(S)
Jeannique Vranken
(J)
Ayla van Ahee
(A)
Hester Bongenaar
(H)
Maylee Smallegange
(M)
Lida Tilet
(L)
Joke de Meris
(J)
Michelle Simons
(M)
Wilma Pellikaan
(W)
Wilma van Wijngaarden
(W)
Kitty Blauwendraat
(K)
Yvonne Drabbe
(Y)
Michelle Sandiman-Lefeber
(M)
Anke Katthöfer
(A)
Eva Ponjee
(E)
Rieke Eilander
(R)
Anja van Loon
(A)
Karin Kraus
(K)
Suze Kooij
(S)
Annemarie Slotboom
(A)
Marieke de Jong
(M)
Friedus van der Minne
(F)
Esther Santegoets
(E)
Leontien Heiligers
(L)
Yvonne Martens
(Y)
Naziha El Ghannouti
(N)
Informations de copyright
© 2022 by American Journal of Neuroradiology.