Committee:
Kesavadas, T (Chair, Director)
Masud, Arif (Director)
Sreenivas, Ramavarapu (Member)
Sutton, Brad (Member)
Decision & Control Systems
FEA-BASED SIMULATION OF BREAST DEFORMATION IN REAL-TIME USING ARTIFICIAL NEURAL NETWORK
Treatment of breast cancer involves two stages: diagnosis (stage one) and treatment (stage two). It is difficult to correlate the imaging results from stage one to stage two because as the patient’s posture changes during treatment, the images captured during diagnosis do not represent the tumor location during the treatment. In the absence of real-time imaging during treatment, the visualization of tumor location is challenging for surgeons.
There are many challenges for breast deformation simulation. For example, material properties are very important to simulate the deformation accurately. The computation speed of simulation will ultimately decide whether the technology is applicable for clinical use or not. Because of hardware limit, achieving real time simulation is often difficult.
This thesis focuses on investigating visualization of breast deformation as the posture of the patient changes. We utilized magnetic resonance imaging (MRI) images of a patient collected prior to this this study to model the deformation. This data was preprocessed to form a 3D reconstructed model of the breast that was used to run a finite element analysis (FEA) simulation. FEA simulates the deformation of breast tissues for different constraints, such as glandular ratio and gravity angle. However, FEA simulation of such deformation can take a few minutes to as much as 40 minutes to complete using an 8 core computer. To obtain real-time visualization, we constructed a neural network (NN) model that takes breast gravity, angle and glandular / fat ratio (breast material) as input to estimate breast deformation for different patient posture. Computation was performed offline. This NN is used to predict the deformation of the breast and provides visualization in real-time (5 ms prediction time).
To further validate our result, we carried out magnetic resonance imaging (MRI) of a breast phantom at several angles (to mimic various patient postures). We also implemented an iterative technique to estimate material properties. This data was used to simulate breast deformations at different posture angles. A similar approach was implemented to build an NN model. Our results show that NN can predict breast deformation under gravity accurately in real-time.
