Korean Journal of Radiology

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

DOI QR Code

Automatic Detection and Classification of Rib Fractures on Thoracic CT Using Convolutional Neural Network: Accuracy and Feasibility

  • Qing-Qing Zhou (Department of Radiology, The Affiliated Jiangning Hospital of Nanjing Medical University) ;
  • Jiashuo Wang (Research Center of Biostatistics and Computational Pharmacy, China Pharmaceutical University) ;
  • Wen Tang (FL 8, Ocean International Center E) ;
  • Zhang-Chun Hu (Department of Radiology, The Affiliated Jiangning Hospital of Nanjing Medical University) ;
  • Zi-Yi Xia (Department of Radiology, The Affiliated Jiangning Hospital of Nanjing Medical University) ;
  • Xue-Song Li (Department of Radiology, The Affiliated Jiangning Hospital of Nanjing Medical University) ;
  • Rongguo Zhang (FL 8, Ocean International Center E) ;
  • Xindao Yin (Department of Radiology, Nanjing First Hospital, Nanjing Medical University) ;
  • Bing Zhang (Department of Radiology, The Affiliated Nanjing Drum Tower Hospital of Nanjing University Medical School) ;
  • Hong Zhang (Department of Radiology, The Affiliated Jiangning Hospital of Nanjing Medical University)
  • Received : 2019.08.31
  • Accepted : 2020.01.21
  • Published : 2020.07.01
Download PDF
⟨ Previous Next ⟩

Abstract

Objective: To evaluate the performance of a convolutional neural network (CNN) model that can automatically detect and classify rib fractures, and output structured reports from computed tomography (CT) images. Materials and Methods: This study included 1079 patients (median age, 55 years; men, 718) from three hospitals, between January 2011 and January 2019, who were divided into a monocentric training set (n = 876; median age, 55 years; men, 582), five multicenter/multiparameter validation sets (n = 173; median age, 59 years; men, 118) with different slice thicknesses and image pixels, and a normal control set (n = 30; median age, 53 years; men, 18). Three classifications (fresh, healing, and old fracture) combined with fracture location (corresponding CT layers) were detected automatically and delivered in a structured report. Precision, recall, and F1-score were selected as metrics to measure the optimum CNN model. Detection/diagnosis time, precision, and sensitivity were employed to compare the diagnostic efficiency of the structured report and that of experienced radiologists. Results: A total of 25054 annotations (fresh fracture, 10089; healing fracture, 10922; old fracture, 4043) were labelled for training (18584) and validation (6470). The detection efficiency was higher for fresh fractures and healing fractures than for old fractures (F1-scores, 0.849, 0.856, 0.770, respectively, p = 0.023 for each), and the robustness of the model was good in the five multicenter/multiparameter validation sets (all mean F1-scores > 0.8 except validation set 5 [512 x 512 pixels; F1-score = 0.757]). The precision of the five radiologists improved from 80.3% to 91.1%, and the sensitivity increased from 62.4% to 86.3% with artificial intelligence-assisted diagnosis. On average, the diagnosis time of the radiologists was reduced by 73.9 seconds. Conclusion: Our CNN model for automatic rib fracture detection could assist radiologists in improving diagnostic efficiency, reducing diagnosis time and radiologists' workload.

Keywords

Acknowledgement

The authors would like to especially acknowledge Qian Gao, MS and Yucai Li, MS (Ocean International Center E, Chaoyang Rd Side Rd, ShiLiPu, Chaoyang Qu, Beijing Shi) for their support in training and testing the convolutional neural network model.

References

  1. Cho SH, Sung YM, Kim MS. Missed rib fractures on evaluation of initial chest CT for trauma patients: pattern analysis and diagnostic value of coronal multiplanar reconstruction images with multidetector row CT. Br J Radiol 2012;85:e845-e850 https://doi.org/10.1259/bjr/28575455
  2. Lin FC, Li RY, Tung YW, Jeng KC, Tsai SC. Morbidity, mortality, associated injuries, and management of traumatic rib fractures. J Chin Med Assoc 2016;79:329-334 https://doi.org/10.1016/j.jcma.2016.01.006
  3. Talbot BS, Gange CP Jr, Chaturvedi A, Klionsky N, Hobbs SK, Chaturvedi A. Traumatic rib injury: patterns, imaging pitfalls, complications, and treatment. Radiographics 2017;37:628-651 https://doi.org/10.1148/rg.2017160100
  4. Liman ST, Kuzucu A, Tastepe AI, Ulasan GN, Topcu S. Chest injury due to blunt trauma. Eur J Cardiothorac Surg 2003;23:374-378 https://doi.org/10.1016/s1010-7940(02)00813-8
  5. Murphy CE 4th, Raja AS, Baumann BM, Medak AJ, Langdorf MI, Nishijima DK, et al. Rib fracture diagnosis in the panscan era. Ann Emerg Med 2017;70:904-909 https://doi.org/10.1016/j.annemergmed.2017.04.011
  6. Langdorf MI, Medak AJ, Hendey GW, Nishijima DK, Mower WR, Raja AS, et al. Prevalence and clinical import of thoracic injury identified by chest computed tomography but not chest radiography in blunt trauma: multicenter prospective cohort study. Ann Emerg Med 2015;66:589-600 https://doi.org/10.1016/j.annemergmed.2015.06.003
  7. Kim J, Kim S, Kim YJ, Kim KG, Park J. Quantitative measurement method for possible rib fractures in chest radiographs. Healthc Inform Res 2013;19:196-204 https://doi.org/10.4258/hir.2013.19.3.196
  8. Kim EY, Yang HJ, Sung YM, Hwang KH, Kim JH, Kim HS. Sternal fracture in the emergency department: diagnostic value of multidetector CT with sagittal and coronal reconstruction images. Eur J Radiol 2012;81:e708-e711 https://doi.org/10.1016/j.ejrad.2011.05.029
  9. Sollmann N, Mei K, Hedderich DM, Maegerlein C, Kopp FK, Loffler MT, et al. Multi-detector CT imaging: impact of virtual tube current reduction and sparse sampling on detection of vertebral fractures. Eur Radiol 2019;29:3606-3616 https://doi.org/10.1007/s00330-019-06090-2
  10. van Laarhoven JJEM, Hietbrink F, Ferree S, Gunning AC, Houwert RM, Verleisdonk EMM, et al. Associated thoracic injury in patients with a clavicle fracture: a retrospective analysis of 1461 polytrauma patients. Eur J Trauma Emerg Surg 2019;45:59-63 https://doi.org/10.1007/s00068-016-0673-6
  11. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016;316:2402-2410 https://doi.org/10.1001/jama.2016.17216
  12. He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K. The practical implementation of artificial intelligence technologies in medicine. Nat Med 2019;25:30-36 https://doi.org/10.1038/s41591-018-0307-0
  13. Choi JS, Han BK, Ko ES, Bae JM, Ko EY, Song SH, et al. Effect of a deep learning framework-based computer-aided diagnosis system on the diagnostic performance of radiologists in differentiating between malignant and benign masses on breast ultrasonography. Korean J Radiol 2019;20:749-758 https://doi.org/10.3348/kjr.2018.0530
  14. Wang H, Cui Z, Chen Y, Avidan M, Abdallah AB, Kronzer A. Predicting hospital readmission via cost-sensitive deep learning. IEEE/ACM Trans Comput Biol Bioinform 2018;15:1968-1978 https://doi.org/10.1109/TCBB.2018.2827029
  15. Onishi Y, Teramoto A, Tsujimoto M, Tsukamoto T, Saito K, Toyama H, et al. Automated pulmonary nodule classification in computed tomography images using a deep convolutional neural network tained by generative adversarial networks. Biomed Res Int 2019;2019:6051939
  16. Gao XW, Qian Y. Prediction of multidrug-resistant TB from CT pulmonary images based on deep learning techniques. Mol Pharm 2018;15:4326-4335
  17. Zura R, Xu ZJ, Della Rocca GJ, Mehta S, Steen RG. When is a fracture not "fresh"? Aligning reimbursement with patient outcome after treatment with low-intensity pulsed ultrasound. J Orthop Trauma 2017;31:248-251 https://doi.org/10.1097/BOT.0000000000000778
  18. Wu X, Jiang Y. [Old fracture]. Zhonghua Wai Ke Za Zhi 2015; 53:460-463
  19. Wootton-Gorges SL, Stein-Wexler R, Walton JW, Rosas AJ, Coulter KP, Rogers KK. Comparison of computed tomography and chest radiography in the detection of rib fractures in abused infants. Child Abuse Negl 2008;32:659-663 https://doi.org/10.1016/j.chiabu.2007.06.011
  20. Arredondo-Gomez E. [Treatment of traumatic clavicular pseudoarthrosis with the Hunec Colchero nail]. Acta Ortop Mex 2007;21:63-68
  21. Huang J, Rathod V, Sun C, Zhu M, Korattikara A, Fathi A, et al. Speed/accuracy trade-offs for modern convolutional object detectors [updated April 2017]. Available at: https://arxiv. org/abs/1611.10012. Accessed June 21, 2019
  22. Redmon J, Farhadi A. YOLOv3: an incremental improvement. Cornell University, 2018. Available at: https://arxiv.org/abs/1804.02767. Accessed June 22, 2019
  23. Setio AAA, Traverso A, de Bel T, Berens MSN, Bogaard CVD, Cerello P, et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challenge. Med Image Anal 2017;42:1-13 https://doi.org/10.1016/j.media.2017.06.015
  24. Wang Y, Yan F, Lu X, Zheng G, Zhang X, Wang C, et al. IILS: intelligent imaging layout system for automatic imaging report standardization and intra-interdisciplinary clinical workflow optimization. EBioMedicine 2019;44:162-181 https://doi.org/10.1016/j.ebiom.2019.05.040
  25. Ren S, He K, Girshick R, Sun J. Faster R-CNN: towards realtime object detection with region proposal networks [updated January 2016]. Cornell University, 2015. Available at: https://arxiv.org/abs/1506.01497. Accessed June 22, 2019