Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Medical Biophysics


Abbas Samani

2nd Supervisor

Ting-Yim Lee


Lung cancer is the most common cause of cancer related death in both men and women. Radiation therapy is widely used for lung cancer treatment. However, this method can be challenging due to respiratory motion. Motion modeling is a popular method for respiratory motion compensation, while biomechanics-based motion models are believed to be more robust and accurate as they are based on the physics of motion. In this study, we aim to develop a biomechanics-based lung tumor tracking algorithm which can be used during External Beam Radiation Therapy (EBRT). An accelerated lung biomechanical model can be used during EBRT only if its boundary conditions (BCs) are defined in a way that they can be updated in real-time. As such, we have developed a lung finite element (FE) model in conjunction with a Neural Networks (NNs) based method for predicting the BCs of the lung model from chest surface motion data.

To develop the lung FE model for tumor motion prediction, thoracic 4D CT images of lung cancer patients were processed to capture the lung and diaphragm geometry, trans-pulmonary pressure, and diaphragm motion. Next, the chest surface motion was obtained through tracking the motion of the ribcage in 4D CT images. This was performed to simulate surface motion data that can be acquired using optical tracking systems. Finally, two feedforward NNs were developed, one for estimating the trans-pulmonary pressure and another for estimating the diaphragm motion from chest surface motion data.

The algorithm development consists of four steps of: 1) Automatic segmentation of the lungs and diaphragm, 2) diaphragm motion modelling using Principal Component Analysis (PCA), 3) Developing the lung FE model, and 4) Using two NNs to estimate the trans-pulmonary pressure values and diaphragm motion from chest surface motion data. The results indicate that the Dice similarity coefficient between actual and simulated tumor volumes ranges from 0.76±0.04 to 0.91±0.01, which is favorable. As such, real-time lung tumor tracking during EBRT using the proposed algorithm is feasible. Hence, further clinical studies involving lung cancer patients to assess the algorithm performance are justified.