University of Twente Student Theses


Estimation of hepatic tumors respiratory motion using learning algorithms and surrogates

Berijanian, M. (2018) Estimation of hepatic tumors respiratory motion using learning algorithms and surrogates.

[img] PDF
Abstract:The respiratory motion in the abdomen is an important source of inaccuracy in clinical applications such as image-guided interventions (e.g. radiotherapy and tumor ablation) and image acquisition (e.g. MRI). The inaccuracies introduced to the treatments or diagnostic tests raise the probabilities of misdiagnosis, incomplete treatment, or destruction of healthy tissues. Among the possible solutions to these problems, much attention has been paid to respiratory motion estimation by means of surrogate signals, in order to compensate the respiratory motion and target the points of interest more accurately. In this study, a correlation between the collected surrogate signals and the liver tumor respiratory motion is obtained using learning-based algorithms. A robotic phantom is developed which simulates the respiratory motion of the liver, the diaphragm, and the abdomen skin in two directions as superior-inferior (SI) and anterior-posterior (AP). The surrogate signals are collected by means of optical markers attached to the abdomen skin and tracked by a digital camera, in addition to an inertial measurement unit (IMU) fixed to the hub of a plastic needle which is inserted into the liver. The liver incorporates a spherical tumor, the displacement of which is measured by an electromagnetic sensor. Using a finite element (FE) model which is developed based on the data collected from the physical phantom as the ground-truth, more surrogate and tumor motion data is generated with different values of parameters such as the tumor size, the tumor location in the liver, and the liver elasticity which differ among patients. Subsequently, a learning algorithm is employed to find a correlation between the tumor respiratory motion and the surrogate signals. A sensitivity analysis is also performed in order to find the effects of the parameters on the tumor respiratory motion. Also, the performance and estimation error of the learning-based model is compared between the estimation results from the measurement data and the results from the simulated data. It is shown that the estimation error of linear regression for the SI and AP directions has been respectively 1.37% and 2.87%, and for quadratic polynomial regression have been 0.76% and 2.41% on the data from the experiments. With the presented phantom design, it is not possible to draw a general conclusion about which surrogate signals have higher correlation with the tumor motion, since it depends completely on the data set. However, by combining all surrogate signals, the estimation error decreases about 0.5-6.5% comparing to using only one of the surrogates. The sensitivity analysis shows that the simulation results are partly correct, and the main difference between the results and the literature information is due to the limitations in the phantom and FEM. Finally, it is discussed that by augmenting the measurement data with the simulated data, the motion estimation error changes from 2.9% to 2%, which is not significant and suggesting that the FEM is a sufficiently good representation of the experimental setup.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Programme:Electrical Engineering MSc (60353)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page