Korean Journal of Radiology

The Korean Society of Radiology (대한영상의학회)
권호 표지
DOI QR코드

DOI QR Code

Fully Automatic Segmentation of Acute Ischemic Lesions on Diffusion-Weighted Imaging Using Convolutional Neural Networks: Comparison with Conventional Algorithms

  • Ilsang Woo (Department of Convergence Medicine, Biomedical Engineering Research Center, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Areum Lee (Department of Convergence Medicine, Biomedical Engineering Research Center, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Seung Chai Jung (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Hyunna Lee (Department of Convergence Medicine, Biomedical Engineering Research Center, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Namkug Kim (Department of Convergence Medicine, Biomedical Engineering Research Center, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Se Jin Cho (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Donghyun Kim (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Jungbin Lee (Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center) ;
  • Leonard Sunwoo (Department of Radiology, Seoul National University Bundang Hospital) ;
  • Dong-Wha Kang (Department of Neurology, University of Ulsan College of Medicine, Asan Medical Center)
  • Received : 2018.09.05
  • Accepted : 2019.03.16
  • Published : 2019.08.01
Download PDF
⟨ Previous Next ⟩

Abstract

Objective: To develop algorithms using convolutional neural networks (CNNs) for automatic segmentation of acute ischemic lesions on diffusion-weighted imaging (DWI) and compare them with conventional algorithms, including a thresholding-based segmentation. Materials and Methods: Between September 2005 and August 2015, 429 patients presenting with acute cerebral ischemia (training:validation:test set = 246:89:94) were retrospectively enrolled in this study, which was performed under Institutional Review Board approval. Ground truth segmentations for acute ischemic lesions on DWI were manually drawn under the consensus of two expert radiologists. CNN algorithms were developed using two-dimensional U-Net with squeeze-and-excitation blocks (U-Net) and a DenseNet with squeeze-and-excitation blocks (DenseNet) with squeeze-and-excitation operations for automatic segmentation of acute ischemic lesions on DWI. The CNN algorithms were compared with conventional algorithms based on DWI and the apparent diffusion coefficient (ADC) signal intensity. The performances of the algorithms were assessed using the Dice index with 5-fold cross-validation. The Dice indices were analyzed according to infarct volumes (< 10 mL, ≥ 10 mL), number of infarcts (≤ 5, 6-10, ≥ 11), and b-value of 1000 (b1000) signal intensities (< 50, 50-100, > 100), time intervals to DWI, and DWI protocols. Results: The CNN algorithms were significantly superior to conventional algorithms (p < 0.001). Dice indices for the CNN algorithms were 0.85 for U-Net and DenseNet and 0.86 for an ensemble of U-Net and DenseNet, while the indices were 0.58 for ADC-b1000 and b1000-ADC and 0.52 for the commercial ADC algorithm. The Dice indices for small and large lesions, respectively, were 0.81 and 0.88 with U-Net, 0.80 and 0.88 with DenseNet, and 0.82 and 0.89 with the ensemble of U-Net and DenseNet. The CNN algorithms showed significant differences in Dice indices according to infarct volumes (p < 0.001). Conclusion: The CNN algorithm for automatic segmentation of acute ischemic lesions on DWI achieved Dice indices greater than or equal to 0.85 and showed superior performance to conventional algorithms.

Keywords

Acknowledgement

The authors gratefully acknowledge technical support from the Medical Imaging and Robotics Laboratory (MIRL), Department of Radiology, Asan Medical Center.

References

  1. Wheeler HM, Mlynash M, Inoue M, Tipirneni A, Liggins J, Zaharchuk G, et al.; DEFUSE 2 Investigators. Early diffusion-weighted imaging and perfusion-weighted imaging lesion volumes forecast final infarct size in DEFUSE 2. Stroke 2013;44:681-685 https://doi.org/10.1161/STROKEAHA.111.000135
  2. Muir KW, Buchan A, von Kummer R, Rother J, Baron JC. Imaging of acute stroke. Lancet Neurol 2006;5:755-768 https://doi.org/10.1016/S1474-4422(06)70545-2
  3. Chen L, Bentley P, Rueckert D. Fully automatic acute ischemic lesion segmentation in DWI using convolutional neural networks. Neuroimage Clin 2017;15:633-643 https://doi.org/10.1016/j.nicl.2017.06.016
  4. Mah YH, Jager R, Kennard C, Husain M, Nachev P. A new method for automated high-dimensional lesion segmentation evaluated in vascular injury and applied to the human occipital lobe. Cortex 2014;56:51-63 https://doi.org/10.1016/j.cortex.2012.12.008
  5. Boldsen JK, Engedal TS, Pedraza S, Cho TH, Thomalla G, Nighoghossian N, et al. Better diffusion segmentation in acute ischemic stroke through automatic tree learning anomaly segmentation. Front Neuroinform 2018;12:21
  6. Straka M, Albers GW, Bammer R. Real-time diffusion-perfusion mismatch analysis in acute stroke. J Magn Reson Imaging 2010;32:1024-1037 https://doi.org/10.1002/jmri.22338
  7. Forbes F, Doyle S, Garcia-Lorenzo D, Barillot C, Dojat M. Adaptive weighted fusion of multiple MR sequences for brain lesion segmentation. 2010 IEEE international symposium on biomedical imaging: from nano to macro;2010 April 14-17;Rotterdam, Netherlands
  8. Charoensuk W, Covavisaruch N, Lerdlum S, Likitjaroen Y. Acute stroke brain infarct segmentation in DWI images. Int J Pharm Med Biol Sci 2015;4:115-122 https://doi.org/10.18178/ijpmbs.4.2.115-122
  9. Prakash KNB, Gupta V, Bilello M, Beauchamp NJ, Nowinski WL. Identification, segmentation, and image property study of acute infarcts in diffusion-weighted images by using a probabilistic neural network and adaptive Gaussian mixture model. Acad Radiol 2006;13:1474-1484 https://doi.org/10.1016/j.acra.2006.09.045
  10. Mohd Saad N, Noor NSM, Abdullah AR, Sobri Muda, Muda AF, Abdul Rahman NNS. Automated stroke lesion detection and diagnosis system. International MultiConference of engineers and computer scientists;2017 March 15-17;Hong Kong, China
  11. Maier O, Wilms M, von der Gablentz J, Kramer UM, Munte TF, Handels H. Extra tree forests for sub-acute ischemic stroke lesion segmentation in MR sequences. J Neurosci Methods 2015;240:89-100 https://doi.org/10.1016/j.jneumeth.2014.11.011
  12. Peng Y, Zhang X, Hu Q. Segmentation of hyper-acute ischemic infarcts from diffusion weighted imaging based on support vector machine. Journal of Computer and Communications 2015;3:152-157 https://doi.org/10.4236/jcc.2015.311024
  13. Nag MK, Koley S, China D, Sadhu AK, Balaji R, Ghosh S, et al. Computer-assisted delineation of cerebral infarct from diffusion-weighted MRI using Gaussian mixture model. Int J Comput Assist Radiol Surg 2017;12:539-552 https://doi.org/10.1007/s11548-017-1520-x
  14. Warach S, Chien D, Li W, Ronthal M, Edelman RR. Fast magnetic resonance diffusion-weighted imaging of acute human stroke. Neurology 1992;42:1717-1723 https://doi.org/10.1212/WNL.42.9.1717
  15. van Everdingen KJ, van der Grond J, Kappelle LJ, Ramos LMP, Mali WPTM. Diffusion-weighted magnetic resonance imaging in acute stroke. Stroke 1998;29:1783-1790 https://doi.org/10.1161/01.STR.29.9.1783
  16. Purushotham A, Campbell BC, Straka M, Mlynash M, Olivot JM, Bammer R, et al. Apparent diffusion coefficient threshold for delineation of ischemic core. Int J Stroke 2015;10:348-353 https://doi.org/10.1111/ijs.12068
  17. Ogata T, Christensen S, Nagakane Y, Ma H, Campbell BC, Churilov L, et al.; EPITHET and DEFUSE Investigators. The effects of alteplase 3 to 6 hours after stroke in the EPITHETDEFUSE combined dataset: post hoc case-control study. Stroke 2013;44:87-93 https://doi.org/10.1161/STROKEAHA.112.668301
  18. Lee JG, Jun S, Cho YW, Lee H, Kim GB, Seo JB, et al. Deep learning in medical imaging: general overview. Korean J Radiol 2017;18:570-584 https://doi.org/10.3348/kjr.2017.18.4.570
  19. Woo I, Lee AR, Lee H, Choi K, Kang DW, Jung SC, et al. Sementic segmentation with squeeze-and-excitation block: application to infarct segmentation in DWI. NIPS 2017: the thirty-first annual conference on Neural Information Processing Systems;2017 December 4-9;Long Beach, CA, USA
  20. Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation. 18th Medical Image Computing and Computer Assisted Interventions (MICCAI);2015 October 5-9;Munich, Germany
  21. Jegou S, Drozdzal M, Vazquez D, Romero A, Bengio Y. The one hundred layers tiramisu: fully convolutional DenseNets for semantic segmentation. IEEE Conference on Computer Vision and Pattern Recognition Workshop (CVPRW);2017 July 21-26; Honolulu, HI, USA
  22. Hu J, Shen L, Sun G. Squeeze-and-excitation networks. IEEE conference on Computer Vision and Pattern Recognition(CVPR);2018 June 18-22; Salt Lake City, UT, USA
  23. Nair V, Hinton GE. Rectified linear units improve restricted boltzmann machines. 27th international conference on international conference on machine learning;2010 June 21-24;Haifa, Israel
  24. Shinohara RT, Sweeney EM, Goldsmith J, Shiee N, Mateen FJ, Calabresi PA, et al.; Australian Imaging Biomarkers Lifestyle Flagship Study of Ageing; Alzheimer's Disease Neuroimaging Initiative. Statistical normalization techniques for magnetic resonance imaging. Neuroimage Clin 2014;6:9-19 https://doi.org/10.1016/j.nicl.2014.08.008
  25. Noh H, Hong S, Han B. Learning deconvolution network for semantic segmentation. IEEE international conference on computer vision (ICCV);2015 December 7-13; Santiago, Chile
  26. Huang G, Liu Z, van der Maaten L, Weinberger KQ. Densely connected convolutional networks. IEEE conference on Computer Vision and Pattern Recognition(CVPR);2017 July 21-26; Honolulu, HI, USA