Traditional deconvolution methods can improve the spatial resolution of photoacoustic computed tomography (PACT) systems but are often sensitive to noise. We propose a novel approach to enhance the resolution of PACT, by modeling the system’s point spread function (PSF) and performing deep-learning-based deconvolution. We train a robust deep learning model without the need for ground truth, using a self-supervised method on a mixed dataset of simulation, phantom, and in vivo data, in combination with various data augmentation techniques. We demonstrate that our deep learning deconvolution achieves superior spatial resolution, image contrast, and artifact suppression, when compared to traditional deconvolution methods.
|