Enhancing large-aperture photoacoustic computed tomography with deep learning: from single finger to human hand imaging

Photoacoustic Computed Tomography (PACT) enables high-resolution imaging by combining optical excitation with ultrasonic detection. We developed a cost-effective large-aperture PACT system using eight linear arrays arranged in a half-ring geometry. While this configuration improves image quality, extended field-of-view imaging is affected by optical shadowing and alignment mismatch. To address these limitations, we applied a U-Net–based deep learning enhancement framework. The network, trained on limited-field PACT data, substantially improved vessel visibility and skin structure continuity. In-vivo human imaging results demonstrate that the proposed approach effectively compensates for large-target imaging artifacts, supporting its applicability to complex anatomical imaging tasks such as peripheral arterial disease assessment.