Deep-Learning Enabled FWI
This project aims to develop two artificial intelligence-driven frameworks to enhance the speed and quality of state-of-the-art ultrasound computed tomography.
The first approach integrates adjoint-tomography theory (ATT) with a generative adversarial network (GAN) to speed up FWI-based USCT by embedding strong priors into the GAN, facilitating rapid and reliable patient screening.
The second DL approach utilizes a physics-guided, cycle-consistent method in both training and application, producing highly detailed reconstructions. This approach reduces reliance on ground truth models, minimizes dependence on initial models, and leverages specialized DL-accelerated computing hardware, thereby lowering false-positive rates and reducing unnecessary testing and biopsies while enhancing early diagnosis and treatment potential.
Sponsor: National Science Foundation (NSF)
Publications: (Refer to the publication page for detailed paper links)
- Coming soon...