An artificial neural network architecture for non-parametric visual odometry in wireless capsule endoscopy

被引:11
作者
Dimas, George [1 ]
Iakovidis, Dimitris K. [1 ]
Karargyris, Alexandros [2 ]
Ciuti, Gastone [3 ]
Koulaouzidis, Anastasios [4 ]
机构
[1] Univ Thessaly, Dept Comp Sci & Biomed Informat, Lamia, Greece
[2] IBM Res, San Jose, CA USA
[3] St Anna Sch Adv Studies, BioRobot Inst, Pisa, Italy
[4] Royal Infirm Edinburgh NHS Trust, Endoscopy Unit, Edinburgh, Midlothian, Scotland
关键词
visual odometry; capsule endoscope localization; neural networks; medical imaging; DIAGNOSIS; MODEL;
D O I
10.1088/1361-6501/aa7ebf
中图分类号
T [工业技术];
学科分类号
08 ;
摘要
Wireless capsule endoscopy is a non-invasive screening procedure of the gastrointestinal (GI) tract performed with an ingestible capsule endoscope (CE) of the size of a large vitamin pill. Such endoscopes are equipped with a usually low-frame-rate color camera which enables the visualization of the GI lumen and the detection of pathologies. The localization of the commercially available CEs is performed in the 3D abdominal space using radio-frequency (RF) triangulation from external sensor arrays, in combination with transit time estimation. State-of-the-art approaches, such as magnetic localization, which have been experimentally proved more accurate than the RF approach, are still at an early stage. Recently, we have demonstrated that CE localization is feasible using solely visual cues and geometric models. However, such approaches depend on camera parameters, many of which are unknown. In this paper the authors propose a novel non-parametric visual odometry (VO) approach to CE localization based on a feed-forward neural network architecture. The effectiveness of this approach in comparison to state-of-the-art geometric VO approaches is validated using a robotic-assisted in vitro experimental setup.
引用
收藏
页数:10
相关论文
共 34 条
[1]  
Ahmed M. T., 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision, P463, DOI 10.1109/ICCV.1999.791257
[2]  
[Anonymous], 1991, DETECTION TRACKING P
[3]  
Bao G., 2013, IEEE International Conference on Electro-Information Technology , EIT, P1, DOI [DOI 10.1109/EIT.2013.6632652, 10.1109/EIT.2013.6632652]
[4]  
Baptista Veronica, 2014, World J Gastrointest Pathophysiol, V5, P523, DOI 10.4291/wjgp.v5.i4.523
[5]   3D Vision by Using Calibration Pattern with Inertial Sensor and RBF Neural Networks [J].
Besdok, Erkan .
SENSORS, 2009, 9 (06) :4572-4585
[6]  
Bouguet J., 2004, Camera Calibration Toolbox for Matlab
[7]  
Corporation G I, 2013, PILLCAM CAPSULE ENDO
[8]   RANDOM SAMPLE CONSENSUS - A PARADIGM FOR MODEL-FITTING WITH APPLICATIONS TO IMAGE-ANALYSIS AND AUTOMATED CARTOGRAPHY [J].
FISCHLER, MA ;
BOLLES, RC .
COMMUNICATIONS OF THE ACM, 1981, 24 (06) :381-395
[9]   UNIVERSAL APPROXIMATION OF AN UNKNOWN MAPPING AND ITS DERIVATIVES USING MULTILAYER FEEDFORWARD NETWORKS [J].
HORNIK, K ;
STINCHCOMBE, M ;
WHITE, H .
NEURAL NETWORKS, 1990, 3 (05) :551-560
[10]  
Hu C, 2004, IEEE INT CONF ROBOT, P4718