Ultrasound Temperature Estimation Using Generative Adversarial Networks
Open Access
- Author:
- Safri, Muayyad
- Area of Honors:
- Computer Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Mohamed Khaled Almekkawy, Thesis Supervisor
Robert Collins, Thesis Honors Advisor - Keywords:
- High-intensity focused ultrasound
Ultrasound Temperature Estimation
Generative adversarial networks
Deep Learning - Abstract:
- High Intensity Focused Ultrasound (HIFU) is a non-invasive ablation technique that has created massive interest in clinical applications such as hyperthermia and thermal therapy. HIFU generates heat at the target region, however, temperature monitoring is essential to avoid damaging healthy tissues in the heated region. Although Magnetic Resonance Imaging (MRI) does provide highly accurate temperature monitoring, ultrasound imaging transducer is more advantageous in temperature estimation due to its portability, low costs, and non-ionizing radiation. Using the physical properties of ultrasound waves, many studies have been conducted for thermal monitoring using HIFU. These classical methods have an unavoidable trade-off between image quality and computational speed. The last decade has seen a significant boost in deep learning and neural networks for all applications due to its high performance while being computationally efficient. We propose to employ a more recent type of neural network, namely a Generative Adversarial Network (GAN), to perform ultrasound temperature estimation. A MATLAB-based HIFU simulator was used to generate the intensity and the corresponding temperature maps for four different tissue combinations, which were used as input data for the GAN. The capabilities of GAN has made it a state-of-the-art neural network for tasks such as image generation, classification and segmentation. This thesis demonstrates the use of GAN in temperature map estimation of the tissue in the region of interest. The promising results show the feasibility of using GAN in clinical-based applications.