DOI QR코드

DOI QR Code

Research on the Performance Optimization of HR-Net for Spinal Region Segmentation in Whole Spine X-ray Images

Whole Spine X-ray 영상에서 척추 영역 분할을 위한 HR-Net 성능 최적화에 관한 연구

  • Han Beom Yu (Department of Radiological, Graduate School, Eulji University) ;
  • Ho Seong Hwang (Machine Intelligence Convergence System, Eulji University ) ;
  • Dong Hyun Kim (Machine Intelligence Convergence System, Eulji University ) ;
  • Hee Jue Oh (Department of Radiological, Graduate School, Eulji University) ;
  • Ho Chul Kim (Department of Radiological, Graduate School, Eulji University)
  • 유한범 (을지대학교 일반대학원 방사선학과) ;
  • 황호성 (을지대학교 인공지능 융합 시스템 연구소) ;
  • 김동현 (을지대학교 인공지능 융합 시스템 연구소) ;
  • 오희주 (을지대학교 일반대학원 방사선학과) ;
  • 김호철 (을지대학교 일반대학원 방사선학과)
  • Received : 2024.04.05
  • Accepted : 2024.08.12
  • Published : 2024.08.31

Abstract

This study enhances AI algorithms for extracting spinal regions from Whole Spine X-rays, aiming for higher accuracy while minimizing learning and detection times. Whole Spine X-rays, critical for diagnosing conditions such as scoliosis and kyphosis, necessitate precise differentiation of spinal contours. The conventional manual methodology encounters challenge due to the overlap of anatomical structures, prompting the integration of AI to overcome these limitations and enhance diagnostic precision. In this study, 1204 AP and 500 LAT Whole Spine X-ray images were meticulously labeled, spanning the third cervical to the fifth lumbar vertebrae. We based our efforts on the HR-Net algorithm, which exhibited the highest accuracy, and proceeded to simplify its network architecture and enhance the block structure for optimization. The optimized HR-Net algorithm demonstrates an improvement, increasing accuracy by 2.98% for the AP dataset and 1.59% for the LAT dataset compared to its original formulation. Additionally, the modification resulted in a substantial reduction in learning time by 70.06% for AP images and 68.43% for LAT images, along with a decrease in detection time by 47.18% for AP and 43.07% for LAT images. The time taken per image for detection was also reduced by 47.09% for AP and 43.07% for LAT images. We suggest that the application of the proposed HR-Net in this study can lead to more accurate and efficient extraction of spinal regions in Whole Spine X-ray images. This can become a crucial tool for medical professionals in the diagnosis and treatment of spinal-related conditions, and it will serve as a foundation for future research aimed at further improving the accuracy and speed of spinal region segmentation.

Keywords

References

  1. Cobb J. Outline for the study of scoliosis. Instructional course lecture. 1948:261-275. 
  2. Stokes IA. Three-dimensional terminology of spinal deformity: a report presented to the Scoliosis Research Society by the Scoliosis Research Society Working Group on 3-D terminology of spinal deformity. Spine. 1994;19(2):236-248. 
  3. Yu HB. Research on application and performance improvement of artificial intelligence algorithm for automating spine segmentation in whole spine X-ray images. Eulji University Graduate School: Daejeon. 2024:1-101. 
  4. Shen D, Wu G, Suk HI. Deep learning in medical image analysis. Annual review of biomedical engineering. 2017;19:221-248. 
  5. Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, Laak J, Ginnken B, Sanchez CI. A survey on deep learning in medical image analysis. Medical image analysis. 2017;42:60-88. 
  6. Lee HY. A Study on Reducing Learning Time of Deep-Learning Algorithm. Hanbat National University Graduate School: Daejeon. 2018:1-48. 
  7. Lee HY, Lee SH. A Study on Reducing Learning Time of Deep-Learning using Network Separation. Journal of Electrical and Electronics Society. 2021;25(2):273-279. 
  8. Yoon BJ, Yu SY. DQN-based Mapless Navigation and learning time reduction algorithm considering moving obstacles. Journal of the Korea Institute of Information & Communication Engineering. 2021;25(3):377-383. 
  9. Sik SB, Jung EY, Yoon CH, Kang KS. An Enhancement of Learning Speed of the Error - Backpropagation Algorithm. Information Processing Society Journal. 1997;4(7):1759-1769. 
  10. Song YS, Seo MS, Kim CW, Kim YH, Yoo BC, Choi HJ, Seo SH, Kang SW, Song MG, Nam DC, Kim DH. AI-Driven Segmentation and Automated Analysis of the Whole Sagittal Spine from X-ray Images for Spinopelvic Parameter Evaluation. Bioengineering. 2023;10(10):1229. 
  11. Sacharisa S, Kartowisastro IH. Enhanced Spine Segmentation in Scoliosis X-ray Images via U-Net. Ingenierie des Systemes d'Information. 2023;28(4):1073-1079. 
  12. Chen Y, Mo Y, Readie A, Ligozio G, Mandal I, Jabbar F, Coroller T, Papiez BW. VertXNet: An Ensemble Method for Vertebrae Segmentation and Identification of Spinal X-Ray. arXiv preprint arXiv:2302.03476. 2023:1-13. 
  13. Kim YT, Jeong TS, Kim YJ, Kim WS, Kim KG, Yee GT. Automatic Spine Segmentation and Parameter Measurement for Radiological Analysis of Whole-Spine Lateral Radiographs Using Deep Learning and Computer Vision. Journal of Digital Imaging. 2023;36:1447-1459. 
  14. Shi W, Xu T, Yang H, Xi Y. Du Y, Li J, Li J. Attention gate based dual-pathway network for vertebra segmentation of X-ray spine images. IEEE Journal of Biomedical and Health Informatics. 2022;26(8):3976-3987. 
  15. Hwang HS, Kim DH, Kim HC. Study on the Application of Artificial Intelligence Model for CT Quality Control. Journal of Biomedical Engineering Research. 2023;44:182-189.  https://doi.org/10.9718/JBER.2023.44.3.182
  16. Hwang HS. A study of CT phantom image quality control evaluation method based on deep learning. Eulji University Graduate School: Daejeon. 2024. 
  17. Peck D. Digital Imaging and Communications in Medicine (DICOM): a practical introduction and survival guide. Soc Nuclear Med. 2009;50(8):1384. 
  18. Bidgood WD, Horii SC, Prior FW, Syckle DE. Understanding and using DICOM, the data interchange standard for biomedical imaging. Journal of the American Medical Informatics Association. 1997;4(3):199-212. 
  19. Horwath JP, Zakharov DN, Megert R, Stach EA. Understanding important features of deep learning models for segmentation of high-resolution transmission electron microscopy images. npj Computational Materials. 2020.;6(1):108. 
  20. Ketkar N, Santana E. Deep learning with Python. Springer. 2017:1-469. 
  21. Everingham M, Eslami SMA, Gool LV, Williams CKI, Winn J, Zisserman A. The pascal visual object classes challenge: A retrospective. International journal of computer vision. 2015;111:98-136. 
  22. Rahman MA, Wang Y. Optimizing intersection-over-union in deep neural networks for image segmentation. in International symposium on visual computing. Springer. 2016:234-244. 
  23. Csurka G, Larlus G, Perronnin F. What is a good evaluation measure for semantic segmentation? in Bmvc. Bristol. 2013:1-11. 
  24. Rezatofighi H, Tsoi N, Gwak JY, Sadeghian A, Reid I, Savarese S. Generalized intersection over union: A metric and a loss for bounding box regression. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019:658-666. 
  25. Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. in Medical Image Computing and Computer-Assisted Intervention-MICCAI. Springer. 2015:234-241. 
  26. Chen LC, Zhu Y, Papandreou G, Schroff F, Adam H. Encoder-decoder with atrous separable convolution for semantic image segmentation. in Proceedings of the European conference on computer vision (ECCV). 2018:801-818. 
  27. Sun K, Xiao B, Liu D, Wang J. Deep high-resolution representation learning for human pose estimation. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2019:5693-5703. 
  28. Zhou B, Khosia A, Lapedriza A, Olivba A, Torralba A. Learning deep features for discriminative localization. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2016:2921-2929. 
  29. Lin MQ, Chen Q, Yan S. Network in network. arXiv preprint arXiv:1312.4400. 2013:1-10. 
  30. Qian Z, Hayes TL, Kafle K, Kanan C. Do we need fully connected output layers in convolutional networks? arXiv preprint arXiv:2004.13587. 2020:1-8. 
  31. Xu B, Wang N, Chen T, Li M. Empirical evaluation of rectified activations in convolutional network. arXiv preprint arXiv:1505.00853. 2015:1-5. 
  32. Hu J, Shen L, Sun G. Squeeze-and-excitation networks. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2018:7132-7141.