The effects of different levels of realism on the training of CNNs with only synthetic images for the semantic segmentation of robotic instruments in a head phantom
Tóm tắt
The manual generation of training data for the semantic segmentation of medical images using deep neural networks is a time-consuming and error-prone task. In this paper, we investigate the effect of different levels of realism on the training of deep neural networks for semantic segmentation of robotic instruments. An interactive virtual-reality environment was developed to generate synthetic images for robot-aided endoscopic surgery. In contrast with earlier works, we use physically based rendering for increased realism. Using a virtual reality simulator that replicates our robotic setup, three synthetic image databases with an increasing level of realism were generated: flat, basic, and realistic (using the physically-based rendering). Each of those databases was used to train 20 instances of a UNet-based semantic-segmentation deep-learning model. The networks trained with only synthetic images were evaluated on the segmentation of 160 endoscopic images of a phantom. The networks were compared using the Dwass–Steel–Critchlow–Fligner nonparametric test. Our results show that the levels of realism increased the mean intersection-over-union (mIoU) of the networks on endoscopic images of a phantom (
$${p}<0.01$$
). The median mIoU values were 0.235 for the flat dataset, 0.458 for the basic, and 0.729 for the realistic. All the networks trained with synthetic images outperformed naive classifiers. Moreover, in an ablation study, we show that the mIoU of physically based rendering is superior to texture mapping (
$${p}<0.01$$
) of the instrument (0.606), the background (0.685), and the background and instruments combined (0.672). Using physical-based rendering to generate synthetic images is an effective approach to improve the training of neural networks for the semantic segmentation of surgical instruments in endoscopic images. Our results show that this strategy can be an essential step in the broad applicability of deep neural networks in semantic segmentation tasks and help bridge the domain gap in machine learning.
Tài liệu tham khảo
Allan M, Shvets A, Kurmann T, Zhang Z, Duggal R, Su Y, Rieke N, Laina I, Kalavakonda N, Bodenstedt S, García-Peraza LC, Li W, Iglovikov V, Luo H, Yang J, Stoyanov D, Maier-Hein L, Speidel S, Azizian M (2019) 2017 robotic instrument segmentation challenge. CoRR. arXiv:1902.06426
Arata J, Fujisawa Y, Nakadate R, Kiguchi K, Harada K, Mitsuishi M, Hashizume M (2019) Compliant four degree-of-freedom manipulator with locally deformable elastic elements for minimally invasive surgery. In: 2019 International conference on robotics and automation (ICRA). IEEE. https://doi.org/10.1109/icra.2019.8793798
Bishop G, Weimer DM (1986) Fast phong shading. ACM SIGGRAPH Comput Graph 20(4):103–106. https://doi.org/10.1145/15886.15897
Budsberg J, Zafar NB, Aldén M (2014) Elastic and plastic deformations with rigid body dynamics. In: ACM SIGGRAPH 2014 talks on-SIGGRAPH14. ACM Press. https://doi.org/10.1145/2614106.2614132
Dolgun A, Demirhan H (2016) Performance of nonparametric multiple comparison tests under heteroscedasticity, dependency, and skewed error distribution. Commun Stat Simul Comput 46(7):5166–5183. https://doi.org/10.1080/03610918.2016.1146761
Douglas CE, Michael FA (1991) On distribution-free multiple comparisons in the one-way analysis of variance. Commun Stati Theory Methods 20(1):127–139. https://doi.org/10.1080/03610929108830487
Hartwig S, Ropinski T (2019) Training object detectors on synthetic images containing reflecting materials. arXiv preprint arXiv:1904.00824
Hinterstoisser S, Pauly O, Heibel H, Marek M, Bokeloh M (2019) An annotation saved is an annotation earned: Using fully synthetic training for object instance detection. CoRR arXiv:1902.09967
Kilgard MJ (2000) A practical and robust bump-mapping technique for todays gpus. Game Dev Conf 2000:1–39
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553):436–444. https://doi.org/10.1038/nature14539
Malangoni MA, Biester TW, Jones AT, Klingensmith ME, Lewis FR (2013) Operative experience of surgery residents: trends and challenges. J Surg Educ 70(6):783–788. https://doi.org/10.1016/j.jsurg.2013.09.015
Marinho MM, Adorno BV, Harada K, Mitsuishi M (2019) Dynamic active constraints for surgical robots using vector-field inequalities. IEEE Trans Robot 35(5):1166–1185. https://doi.org/10.1109/tro.2019.2920078
Marinho MM, Harada K, Morita A, Mitsuishi M (2019) Smartarm: integration and validation of a versatile surgical robotic system for constrained workspaces. Int J Med Robot Comput Assist Surg. https://doi.org/10.1002/rcs.2053
Masuda T, Kanako H, Adachi S, Arai F, Omata S, Morita A, kin T, Saito N, Yamashita J, Chinzei K, Haswgawa A, Fukuda T. (2018) Patients simulator for transsphenoidal surgery. In: 2018 International symposium on micro-nanomechatronics and human science (MHS). IEEE https://doi.org/10.1109/mhs.2018.8886922
Peddie J (2019) Work flow and material standards. In: Ray tracing: a tool for all, pp 65–90. Springer. https://doi.org/10.1007/978-3-030-17490-3_5
Pharr M, Humphreys G, Jakob W (2016) Physically based rendering. Elsevier LTD, Oxford. https://www.ebook.de/de/product/25867811/matt_pharr_greg_humphreys_wenzel_jakob_physically_based_rendering.html
Ren S, He K, Girshick R, Sun J (2015) Faster r-cnn: towards real-time object detection with region proposal networks. In: Advances in neural information processing systems, pp. 91–99. http://papers.nips.cc/paper/5638-faster-r-cnn-towards-real-time-object-detection-with-region-proposal-networks
Reznick RK, MacRae H (2006) Teaching surgical skills: changes in the wind. N Engl J Med 355(25):2664–2669. https://doi.org/10.1056/nejmra054785
Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. In: Lecture notes in computer science, pp 234–241. Springer. https://doi.org/10.1007/978-3-319-24574-4_28
Shelhamer E, Long J, Darrell T (2017) Fully convolutional networks for semantic segmentation. IEEE Trans Pattern Anal Mach Intel 39(4):640–651. https://doi.org/10.1109/tpami.2016.2572683
Tobin J, Fong R, Ray A, Schneider J, Zaremba W, Abbeel P (2017) Domain randomization for transferring deep neural networks from simulation to the real world. In: 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE. https://doi.org/10.1109/iros.2017.8202133
Torrance KE, Sparrow EM (1967) Theory for off-specular reflection from roughened surfaces. J Opt Soc Am 57(9):1105. https://doi.org/10.1364/josa.57.001105
van der Sluis PC, Ruurda JP, van der Horst S, Goense L, van Hillegersberg R (2018) Learning curve for robot-assisted minimally invasive thoracoscopic esophagectomy: results from 312 cases. Ann Thorac Surg 106(1):264–271. https://doi.org/10.1016/j.athoracsur.2018.01.038
Zisimopoulos O, Flouty E, Stacey M, Muscroft S, Giataganas P, Nehme J, Chow A, Stoyanov D (2017) Can surgical simulation be used to train detection and classification of neural networks? Healthc Technol Lett 4(5):216–222. https://doi.org/10.1049/htl.2017.0064