Publication Date

2018-05-09

Availability

Embargoed

Embargo Period

2020-05-08

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PHD)

Department

Biomedical Engineering (Engineering)

Date of Defense

2018-04-26

First Committee Member

Weizhao Zhao

Second Committee Member

Xiaodong Wu

Third Committee Member

Ivaylo Mihaylov

Fourth Committee Member

Edward Dauer

Fifth Committee Member

Yidong Yang

Sixth Committee Member

Nelson Salas

Abstract

For the Cyberknife radiation therapy treatment modality, the gold standard of real-time tracking of abdominal tumors requires the use of fiducial markers. The use of fiducial markers has not only reduced the dose to the organs-at-risk (OARs) but has also improved the accuracy with which tumors influenced by motion are tracked during treatment delivery. However, there are some drawbacks when using fiducial markers such as infection, pain at the injection site, treatment delay and cost. These complications were a catalyst for many research groups trying to determine the feasibility of using the diaphragm as a motion surrogate. The main limitations for these existing approaches are 1. the imaging modalities used for tumor monitoring, in their studies, are not currently available for the Cyberknife treatment modality and 2. their reported tracking errors are higher than that of fiducial-based real-time tracking. We hypothesize that the two-dimensional (2D) location of the lung-diagram border can be used to determine the three-dimensional (3D) location of the tumor with improved accuracy in comparison to previous studies using existing kV imagers and artificial neural networks (ANN). For our proof of concept study, we first performed a simulation study with the use of a digital phantom known as the 4D extended cardiac-torso (XCAT) phantom. Subsequently, using real patient 4DCT data, a validation study was performed. The proposed methodology utilizes an abdominal 4DCT dataset containing multiple phases of one breathing cycle. One set of digitally reconstructed radiograph’s (DRR) images was generated at ±45° for each phase. On each DRR, an outline of the lung-diaphragm border was detected using an edge detection algorithm. The tumor volume’s gravity center was identified for each phase of the breathing cycle which serves as the measured center of the volume. Using a simple 2-layer ANN architecture, correlation models correlating the lung-diaphragm border’s location with the corresponding 3D location of the tumor volume, were compared. The testing of these models was done using the mean root squared error (MRSE) values and the leave-one-out (LOO) validation technique. An overall tracking error of < 1 mm (0.17 - 1.38 mm) was observed for a variety of regular and irregular breathing patterns of the phantom. For the real patient data, tracking errors of ~2mm (0.97 - 3.87 mm) were observed. The reconstruction error was shown to have the most influence on the tracking error as the dataset with the smallest and largest reconstruction error produced tracking errors of 0.97 mm and 3.87 mm respectively. Other factors influencing the tracking error include DRR projection angle, superior-inferior distance between the lung-diaphragm border and tracking volume and regular vs. irregular motion patterns. This proof of concept study shows the feasibility of accurately predicting the tumor volume’s position with the use of existing kV imagers and has the potential to eliminate fiducial markers in the tracking of liver or other abdominal tumors.

Keywords

Real-Time Tracking; Artficial Neural Networks; Liver; Diaphragm; Fiducial-Less; Radiation Therapy

Available for download on Friday, May 08, 2020

Share

COinS