基於內視鏡影像序列之手術器械辨識與追蹤

Surgical Instrument Recognition and Tracking Using Endoscopic Image Sequences

陳俊儒(Chun-Ju Chen)

指導教授 ： 宋開泰

國立交通大學/電機學院/電機與控制工程系所/碩士(2013年)

https://doi.org/10.6842/NCTU.2013.00553

摘要

本論文主旨在研究內視鏡扶持機器人之影像追蹤系統。內視鏡裝置於扶持機器人上，本系統會透過內視鏡的影像即時偵測手術器械，並根據器械在影像上的位置自主調整內視鏡扶持機器人運動，帶動內視鏡的移動以提供適當的影像視野。在影像辨識設計部份，本論文提出基於Spiking Neural Network(SNN)演算法，利用手術器械之紋理和幾何等自然特徵來辨識內視鏡影像中之手術器械。透過資料訓練後，類神經網路辨識器不容易受光線變化所影響；器械的大小變化或形變等辨識問題也能被克服。本論文結合Region of interest及Kalman filter估測影像畫面中器械之位置以提升辨識的效率。在手術器械追蹤控制方面，考慮到內視鏡對器械的追蹤太過敏感會導致手術中螢幕影像畫面過度晃動而干擾醫師，我們提出「緩衝區」的設計，以進行手術器械之追蹤控制。如此一來，內視鏡機器人在追蹤器械的同時，也能提供穩定的影像畫面。所發展之方法先以內視鏡影像驗證器械之辨識率可達91%以上; 進而在華陀機器人上進行影像追蹤實驗，成功驗證本論文所發展方法之有效性。

關鍵字

手術器械追蹤 ； 微創手術 ； 視覺伺服 ； spiking 神經網路

並列摘要

The objective of this study is to design an image tracking algorithm for the endoscopic system in Minimally Invasive Surgery (MIS). The endoscopic robot autonomously adjusts its pose according to the position of the instruments in image plane, and moves the endoscope to provide a suitable field of view. A method is proposed to identify the tip of instruments without using extra artificial markers. We suggest to use texture and geometric features of laparoscopic instruments and to adopt the spiking neural network approach for object detection. Affection of light change can be reduced. The size change problem and deformation of the instrument can be handled by the neural network. To enhance tracking performance, we further employ region of interest(ROI) and Kalman filter to the neuro-based tracker. For the tracking control of surgical instrument, we propose to set a buffer zone in the center of the image frame to avoid redundant movement of the camera. In this way, the endoscopic system provides a stable view while the robot is tracking surgical instruments. By using endoscopic images, a recognition rate above 91% has been achieved for surgical instruments. Practical experiments on Huatuo robot further validate the effectiveness of the developed image tracking methods.

並列關鍵字

instrument tracking ； minimally invasive surgery ； visual servoing ； spiking neural network

參考文獻

[1] F. Tendick and M. C. Cavusoglu, “Human-Machine Interfaces for Minimally Invasive Surgery,” in Proc. of IEEE International Conference on Engineering in Medicine and Biology Society, Chicago, America, Oct 30- Nov 5, 1997, pp.2771-2776.

[7] “Freehand Ltd,” available: http://www.freehandsurgeon.com/

[10] L. Bouarfa, O. Akman, A. Schneider, Pieter P. Jonker and J. Dankelman, “In-vivo Real-time Tracking of Surgical Instruments in Endoscopic Video,” Minimally Invasive Therapy & Allied Technologies, vol.21, no.3, pp.129-134, 2012.

[12] S. Speidel, J. Benzko, S. Krappe, G. Sudra, P. Azad, B. Peter, M. Stich, C. Gutt and R.Dillmann, “Automatic Classification of Minimally Invasive Instruments based on Endoscopic Image Sequences,” in Proc. of SPIE on Medical Imaging, vol. 7261, pp. 1-8, 2008.

[14] V. Sa-Ing, S. S. Thongvigitmanee, C. Wilasrusmee and J.Suthakorn, “Object Tracking for Laparoscopic Surgery Using the Adaptive Mean-Shift Kalman Algorithm,” International Journal of Machine Learning and Computing, Vol. 1, No. 5, pp.441-447, 2011.