Integration of wave physics and machine learning to enable transcranial photoacoustic tomography

Mark Anastasio, University of Illinois at Urbana-Champaign

Usage Details

Neuroimaging technologies are playing an increasingly important role in the initial detection and subsequent monitoring of a wide range of brain diseases and injuries in children [1]. Such applications include detection and management of traumatic brain injury and tumors, the assessment of stroke risk in children with sickle cell disease, and imaging the preterm children brain, to name only a few [2, 3]. In this project, we address the computational challenges of a 3D photoacoustic computed tomography (PACT) neuroimaging modality that has the promise to provide high-resolution, safe and rapid functional brain imaging [4-8]. The goal of this project is to develop novel 3D PACT image reconstruction methods that integrate wave physics and machine learning. To achieve this, we will perform large scale high-fidelity acoustic-elastic wave propagation using our massively parallel multi-GPUs implementation of a forward-adjoint operator pair based on a finite-difference time-domain (FDTD) discretization [9, 10]. Access to Blue Waters resources will allow us to construct a large dataset of human brain numerical phantoms and corresponding PACT measurements for the training of deep learning enhanced image reconstruction methods.