Korean Journal of Radiology

  • 제25권4호
  • /
  • Pages.363-373
  • /
  • 2024
  • /
  • 1229-6929(pISSN)
  • /
  • 2005-8330(eISSN)
대한영상의학회 (The Korean Society of Radiology)
권호 표지
DOI QR코드

DOI QR Code

Automated Detection and Segmentation of Bone Metastases on Spine MRI Using U-Net: A Multicenter Study

  • Dong Hyun Kim (Department of Radiology, SMG-SNU Boramae Medical Center) ;
  • Jiwoon Seo (Department of Radiology, SMG-SNU Boramae Medical Center) ;
  • Ji Hyun Lee (Department of Radiology, SMG-SNU Boramae Medical Center) ;
  • Eun-Tae Jeon (Department of Radiology, SMG-SNU Boramae Medical Center) ;
  • DongYoung Jeong (DEEPNOID Inc.) ;
  • Hee Dong Chae (College of Medicine, Seoul National University) ;
  • Eugene Lee (College of Medicine, Seoul National University) ;
  • Ji Hee Kang (Department of Radiology, Konkuk University Medical Center) ;
  • Yoon-Hee Choi (Department of Physical Medicine and Rehabilitation, Soonchunhyang University Seoul Hospital) ;
  • Hyo Jin Kim (Department of Radiology, SMG-SNU Boramae Medical Center) ;
  • Jee Won Chai (Department of Radiology, SMG-SNU Boramae Medical Center)
  • 투고 : 2022.10.11
  • 심사 : 2024.01.13
  • 발행 : 2024.04.01
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

Objective: To develop and evaluate a deep learning model for automated segmentation and detection of bone metastasis on spinal MRI. Materials and Methods: We included whole spine MRI scans of adult patients with bone metastasis: 662 MRI series from 302 patients (63.5 ± 11.5 years; male:female, 151:151) from three study centers obtained between January 2015 and August 2021 for training and internal testing (random split into 536 and 126 series, respectively) and 49 MRI series from 20 patients (65.9 ± 11.5 years; male:female, 11:9) from another center obtained between January 2018 and August 2020 for external testing. Three sagittal MRI sequences, including non-contrast T1-weighted image (T1), contrast-enhanced T1-weighted Dixon fat-only image (FO), and contrast-enhanced fat-suppressed T1-weighted image (CE), were used. Seven models trained using the 2D and 3D U-Nets were developed with different combinations (T1, FO, CE, T1 + FO, T1 + CE, FO + CE, and T1 + FO + CE). The segmentation performance was evaluated using Dice coefficient, pixel-wise recall, and pixel-wise precision. The detection performance was analyzed using per-lesion sensitivity and a free-response receiver operating characteristic curve. The performance of the model was compared with that of five radiologists using the external test set. Results: The 2D U-Net T1 + CE model exhibited superior segmentation performance in the external test compared to the other models, with a Dice coefficient of 0.699 and pixel-wise recall of 0.653. The T1 + CE model achieved per-lesion sensitivities of 0.828 (497/600) and 0.857 (150/175) for metastases in the internal and external tests, respectively. The radiologists demonstrated a mean per-lesion sensitivity of 0.746 and a mean per-lesion positive predictive value of 0.701 in the external test. Conclusion: The deep learning models proposed for automated segmentation and detection of bone metastases on spinal MRI demonstrated high diagnostic performance.

키워드

과제정보

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MSIT) (No. 2019R1C1C100904413).

참고문헌

  1. Coleman RE. Metastatic bone disease: clinical features, pathophysiology and treatment strategies. Cancer Treat Rev 2001;27:165-176 
  2. Levack P, Graham J, Collie D, Grant R, Kidd J, Kunkler I, et al. Don't wait for a sensory level--listen to the symptoms: a prospective audit of the delays in diagnosis of malignant cord compression. Clin Oncol (R Coll Radiol) 2002;14:472-480 
  3. Haanen JB, Robert C. Immune checkpoint inhibitors. Prog Tumor Res 2015;42:55-66 
  4. Weber KL, Lewis VO, Randall RL, Lee AK, Springfield D. An approach to the management of the patient with metastatic bone disease. Instr Course Lect 2004;53:663-676 
  5. Ait Skourt B, El Hassani A, Majda A. Lung CT image segmentation using deep neural networks. Procedia Comput Sci 2018;127:109-113 
  6. Chen H, Dou Q, Yu L, Qin J, Heng PA. VoxResNet: deep voxelwise residual networks for brain segmentation from 3D MR images. Neuroimage 2018;170:446-455 
  7. Lee H, Tajmir S, Lee J, Zissen M, Yeshiwas BA, Alkasab TK, et al. Fully automated deep learning system for bone age assessment. J Digit Imaging 2017;30:427-441 
  8. Zhao Z, Pi Y, Jiang L, Xiang Y, Wei J, Yang P, et al. Deep neural network based artificial intelligence assisted diagnosis of bone scintigraphy for cancer bone metastasis. Sci Rep 2020;10:17046 
  9. Sim Y, Chung MJ, Kotter E, Yune S, Kim M, Do S, et al. Deep convolutional neural network-based software improves radiologist detection of malignant lung nodules on chest radiographs. Radiology 2020;294:199-209 
  10. Kim J, Min JH, Kim SK, Shin SY, Lee MW. Detection of hepatocellular carcinoma in contrast-enhanced magnetic resonance imaging using deep learning classifier: a multicenter retrospective study. Sci Rep 2020;10:9458 
  11. Hille G, Dunnwald M, Becker M, Steffen J, Saalfeld S, Tonnies K. Segmentation of vertebral metastases in MRI using an U-Net like convolutional neural network. In: Handels H, Deserno T, Maier A, Maier-Hein K, Palm C, Tolxdorff T, eds. Bildverarbeitung fur die medizin 2019. Wiesbaden: Springer Vieweg, 2019:31-36 
  12. Grovik E, Yi D, Iv M, Tong E, Rubin D, Zaharchuk G. Deep learning enables automatic detection and segmentation of brain metastases on multisequence MRI. J Magn Reson Imaging 2020;51:175-182 
  13. Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: Navab N, Hornegger J, Wells W, Frangi A, eds. Medical image computing and computer-assisted intervention-MICCAI 2015. Cham: Springer, 2015:234-241 
  14. Cicek O, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O. 3D U-Net: learning dense volumetric segmentation from sparse annotation. In: Ourselin S, Joskowicz L, Sabuncu M, Unal G, Wells W, eds. Medical image computing and computer-assisted intervention-MICCAI 2016. Cham: Springer, 2016:424-432 
  15. Dice LR. Measures of the amount of ecologic association between species. Ecology 1945;26:297-302 
  16. Sorensen TJ. A method of establishing groups of equal amplitude in plant sociology based on similarity of species content and its application to analyses of the vegetation on Danish commons [dissertation]. Copenhagen: I Kommission hos E. Munksgaard, 1948 
  17. Wang J, Fang Z, Lang N, Yuan H, Su MY, Baldi P. A multi-resolution approach for spinal metastasis detection using deep Siamese neural networks. Comput Biol Med 2017;84:137-146