H van Voorst1,2, P R Konduri3,2, L M van Poppel3,2, W van der Steen4,5, P M van der Sluijs4,5, E M H Slot6, B J Emmer3, W H van Zwam7, Y B W E M Roos8, C B L M Majoie3, G Zaharchuk9, M W A Caan2, H A Marquering. 1. From the Departments of Radiology and Nuclear Medicine (H.v.V., P.R.K., L.M.v.P., B.J.E., C.B.L.M.M., H.A.M.) h.vanvoorst@amsterdamumc.nl. 2. Biomedical Engineering and Physics (H.v.V., P.R.K., L.M.v.P., M.W.A.C., H.A.M.). 3. From the Departments of Radiology and Nuclear Medicine (H.v.V., P.R.K., L.M.v.P., B.J.E., C.B.L.M.M., H.A.M.). 4. Departments of Neurology (W.v.d.S., P.M.v.d.S.). 5. Radiology and Nuclear Medicine (W.v.d.S., P.M.v.d.S.), Erasmus University Medical Center, Rotterdam, the Netherlands. 6. Department of Neurology and Neurosurgery (E.M.H.S.), University Medical Center Utrecht, Utrecht, the Netherlands. 7. Department of Radiology and Nuclear Medicine (W.H.v.Z.), Maastricht University Medical Center, Maastricht, the Netherlands. 8. Neurology (Y.B.W.E.M.R.), Faculty of Medicine, Amsterdam University Medical Centers, University of Amsterdam, Amsterdam, the Netherlands. 9. Department of Radiology (G.Z.), Stanford University, Stanford, California.
Abstract
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.
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.
Authors: Stefan Klein; Marius Staring; Keelin Murphy; Max A Viergever; Josien P W Pluim Journal: IEEE Trans Med Imaging Date: 2009-11-17 Impact factor: 10.048
Authors: J Claude Hemphill; Steven M Greenberg; Craig S Anderson; Kyra Becker; Bernard R Bendok; Mary Cushman; Gordon L Fung; Joshua N Goldstein; R Loch Macdonald; Pamela H Mitchell; Phillip A Scott; Magdy H Selim; Daniel Woo Journal: Stroke Date: 2015-05-28 Impact factor: 7.914
Authors: Anna M M Boers; Ivo G H Jansen; Henk A Marquering; Charles B L M Majoie; Ludo F M Beenen; Thomas G Devlin; Luis San Roman; Ji Hoe Heo; Marc Ribó; Scott Brown; Mohammed A Almekhlafi; David S Liebeskind; Jeanne Teitelbaum; Hester F Lingsma; Wim H van Zwam; Patricia Cuadras; Richard du Mesnil de Rochemont; Marine Beaumont; Martin M Brown; Albert J Yoo; Robert J van Oostenbrugge; Bijoy K Menon; Geoffrey A Donnan; Jean Louis Mas; Yvo B W E M Roos; Catherine Oppenheim; Aad van der Lugt; Richard J Dowling; Michael D Hill; Antoni Davalos; Thierry Moulin; Nelly Agrinier; Andrew M Demchuk; Demetrius K Lopes; Lucia Aja Rodríguez; Diederik W J Dippel; Bruce C V Campbell; Peter J Mitchell; Fahad S Al-Ajlan; Tudor G Jovin; Jeremy Madigan; Gregory W Albers; Sebastien Soize; Francis Guillemin; Vivek K Reddy; Serge Bracard; Jordi Blasco; Keith W Muir; Raul G Nogueira; Phil M White; Mayank Goyal; Stephen M Davis Journal: J Neurointerv Surg Date: 2018-04-07 Impact factor: 5.836
Authors: Kilian M Treurniet; Natalie E LeCouffe; Manon Kappelhof; Yvo B W E M Roos; Charles B L M Majoie; Bart J Emmer; Adriaan C G M van Es; Jelis Boiten; Geert J Lycklama; Koos Keizer; Lonneke S F Yo; Hester F Lingsma; Wim H van Zwam; Inger de Ridder; Robert J van Oostenbrugge; Aad van der Lugt; Diederik W J Dippel; Jonathan M Coutinho Journal: Trials Date: 2021-02-15 Impact factor: 2.279
Authors: F A V Anne Pirson; Wouter H Hinsenveld; Robert-Jan B Goldhoorn; Julie Staals; Inger R de Ridder; Wim H van Zwam; Marianne A A van Walderveen; Geert J Lycklama À Nijeholt; Maarten Uyttenboogaart; Wouter J Schonewille; Aad van der Lugt; Diederik W J Dippel; Yvo B W E M Roos; Charles B L M Majoie; Robert J van Oostenbrugge Journal: Trials Date: 2021-02-24 Impact factor: 2.279