Automatic CT image segmentation of maxillary sinus based on VGG network and improved V-Net

Springer Science and Business Media LLC - Tập 15 - Trang 1457-1465 - 2020
Jiangchang Xu1, Shiming Wang1, Zijie Zhou2, Jiannan Liu2, Xiaoyi Jiang3, Xiaojun Chen1
1Institute of Biomedical Manufacturing and Life Quality Engineering, State Key Laboratory of Mechanical System and Vibration, School of Mechanical Engineering, Shanghai Jiao Tong University, Minhang District, Shanghai, China
2Shanghai Ninth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China
3Faculty of Mathematics and Computer Science, University of Münster, Münster, Germany

Tóm tắt

The analysis of the maxillary sinus (MS) can provide an assessment for many clinical diagnoses, so accurate CT image segmentation of the MS is essential. However, common segmentation methods are mainly done by experienced doctors manually, and there are some challenges such as low efficiency and precision. As for automatic methods, the initial seed points and adjustment of various parameters are required, which will affect the segmentation efficiency. Thus, accurate, efficient, and automatic segmentation method of MS is critical to promote the clinical application. This paper proposed an automatic CT image segmentation method of MS based on VGG network and improved V-Net. The VGG network was established to classify CT slices, which can avoid the failure of CT slice segmentation without MS. Then, we proposed the improved V-Net based on edge supervision for segmenting MS regions more effectively. The edge loss was integrated into the loss of the improved V-Net, which could reduce region misjudgment and improve the automatic segmentation performance. For the classification of CT slices with MS and without MS, the VGG network had a classification accuracy of 97.04 ± 2.03%. In the segmentation, our method obtained a better result, in which the segmentation Dice reached 94.40 ± 2.07%, the Iou (intersection over union) was 90.05 ± 3.26%, and the precision was 94.72 ± 2.64%. Compared with U-Net and V-Net, it reduced region misjudgment significantly and improved the segmentation accuracy. By analyzing the error map of 3D reconstruction, it was mainly distributed in ± 1 mm, which demonstrated that our result was quite close to the ground truth. The segmentation of the MS can be realized efficiently, accurately, and automatically by our method. Meanwhile, it not only has a better segmentation result, but also improves the doctor’s work efficiency, which will have significant impact on clinical applications in the future.

Tài liệu tham khảo

Giacomini G, Pavan ALM, Altemani JMC, Duarte SB, Fortaleza CMCB, Miranda JRA, De Pina DR (2018) Computed tomography-based volumetric tool for standardized measurement of the maxillary sinus. PLoS ONE 13(1):e0190770 Aksoy EA, Özden SU, Karaarslan E, Ünal ÖF (2014) Tanyeri H (2014) Reliability of high-pitch ultra-low-dose paranasal sinus computed tomography for evaluating paranasal sinus anatomy and sinus disease. J Craniofac Surg 25(5):1801–1804 Pallanch JF, Yu L, Delone D, Robb R, Holmes DR 3rd, Camp J, Edwards P, McCollough CH, Ponikau J, Dearking AC, Lane J, Primak A, Shinkle A, Hagan J, Frigas E, Ocel JJ, Tombers N, Siwani R, Orme NM, Reed KB, Jerath N, Dhillon R, Kita H (2013) Three-dimensional volumetric computed tomographic scoring as an objective outcome measure for chronic rhinosinusitis: clinical correlations and comparison to Lund-Mackay scoring. Int Forum Allergy Rhinol 3(12):963–972 Berberi A, Bouserhal L, Nader N, Assaf RB, Nassif NB, Bouserhal J, Salameh Z (2015) Evaluation of three-dimensional volumetric changes after sinus floor augmentation with mineralized cortical bone allograft. J Maxillofac Oral Surg 14(3):624–629 Starch-Jensen T, Jensen JD (2017) Maxillary sinus floor augmentation: a review of selected treatment modalities. J Oral Maxillofac Res 8(3):e3 Demir UL, Akca ME, Ozpar R, Albayrak C (2015) Anatomical correlation between existence of concha bullosa. Surg Radiol Anat 37:1093–1098 Andersen TN, Darvann TA, Murakami S, Larsen P, Senda Y, Bilde A, Buchwald C, Kreiborg S (2018) Accuracy and precision of manual segmentation of the maxillary sinus in MR images-a method study. Br J Radiol 91(1085):20170663 Gomes AF, Gamba TO, Yamasaki MC, Groppo FC, Neto FH, Possobon RF (2018) Development and validation of a formula based on maxillary sinus measurements as a tool for sex estimation: a cone beam computed tomography study. Int J Legal Med 133:1241–1249 Shi H, Scarfe WC, Farman AG (2006) Maxillary sinus 3D segmentation and reconstruction from cone beam CT data sets. Int J Comput Assist Radiol Surg 1:83–89 Jinda-apiraksa A, Ongy SH, Hiewy LT, Foongz KWC, Kondo T (2009) A segmentation technique for maxillary sinus using the 3-D level set method. In: Proceedings of the IEEE region 10 conference, pp 1–6 Shelhamer E, Long J, Darrell T (2015) Fully convolutional networks for semantic segmentation. IEEE Trans Pattern Anal Mach Intell 39(4):640–651 Ronneberger O, Fischer P, Brox T (2015) U-Net: convolutional networks for biomedical image segmentation. In: Proceedings of the international conference on medical image computing and computer-assisted intervention, Springer, Cham, pp 234–241 Milletari F, Navab N, Ahmadi SA (2016) V-Net: fully convolutional neural networks for volumetric medical image segmentation. In: Proceedings of the IEEE fourth international conference on 3D vision, pp 565–571 Tang H, Chen X, Liu Y, Lu Z, You J, Yang M, Yao S, Zhao G, Xu Y, Chen T, Liu Y, Xie X (2019) Clinically applicable deep learning framework for organs at risk delineation in CT images. Nat Mach Intell 1:480–491 Xu J, Jing M, Wang S, Yang C, Chen X (2019) A review of medical image detection for cancers in digestive system based on artificial intelligence. Expert Rev Med Devices 16(10):877–889 Simonyan K, Zisserman A (2015) Very deep convolutional networks for large-scale image recognition. In: Proceedings of the international conference on learning representations, pp 1–14 Xu Z, Wu Z, and Feng J (2018) CFUN: combining faster R-CNN and U-Net network for efficient whole heart segmentation. arXiv preprint, arXiv:1812.04914 Zimmermann RS, Siems JN (2019) Faster training of Mask R-CNN by focusing on instance boundaries. Comput Vis Image Underst 188:102795 Kervadec H, Bouchtiba J, Desrosiers C, Granger É, Dolz J, Ayed IB (2019) Boundary loss for highly unbalanced segmentation. In: Proceedings of the international conference on medical imaging with deep learning (MIDL), pp 285–296 Ioffe S, Szegedy C (2015) Batch normalization: accelerating deep network training by reducing internal covariate shift. In: Proceedings of the international conference on international conference on machine learning, pp 448–456 Ham S, Lee A, Park J, Byeon Y, Lee S, Bae M, Kim N (2018) Multi-structure segmentation of hard tissues, maxillary sinus, mandible, mandibular canals in cone beam CT of head and neck with 3D U-Net. In: Proceedings of the international conference on medical imaging with deep learning (MIDL), pp 1–3