AI-assisted Model-guided Image Reconstruction Methods for Photoacoustic and Ultrasound Computed Tomography

Abstract

Photoacoustic computed tomography (PACT) and ultrasound computed tomography (USCT) are two emerging medical modalities that hold great promise for both pre-clinical and clinical imaging applications, including cancer diagnosis and management. PACT is a hybrid imaging modality that combines benefits from both optical and acoustic imaging to provide high contrast images of biological absorbers. USCT is an ultrasound imaging modality that provides high-resolution quantitative images of acoustic properties utilizing accurate physics models. Both PACT and USCT can be applied for monitoring or diagnosis of cancer and are non-invasive and radiation free. However, both imaging modalities are in the early stages of development and suffer fundamental difficulties that prevent their widespread adoption. To further develop PACT and USCT, this dissertation pursues three specific aims: 1) Enable high-resolution 4D PACT with scalable neural field representations; 2) Demonstrate in-vivo estimation of tumor perfusion rates using 5D DCE-PACT; 3) Accelerate USCT reconstruction with learned and hybrid FWI methods. Achieving these aims will result in computationally efficient image reconstruction methods for PACT and USCT and allow them to transfer into medical applications. Widespread use of PACT and USCT will then provide important tools for medical experts to monitor the progression of cancer and its response to treatments.