Multi-angle phase aberration correction for holographic tomography by dual-output U-Net.

Holographic tomography based on multi-angle reconstruction of complex amplitude field distributions can retrieve the internal refractive-index distribution of the object under investigation. However, its spatial resolution and imaging quality are often limited by optical aberrations. Traditional aberration correction methods face limitations such as high cost and vulnerability to abnormal conditions. To address these challenges, we propose a dual-output-branch three-dimensional convolutional neural network that optimizes the input complex-amplitude distributions acquired under multi-angle illumination via end-to-end learning, thereby enabling accurate aberration correction for tomographic imaging. Taking multi-angle phase maps and masks as inputs, the network performs full-angle aberration correction via the complementary outputs of Zernike coefficients and aberration phase maps. Experiments demonstrate that the refractive index distribution reconstructed by our method achieves an SSIM of 0.9929 and effectively preserves microscopic details. Therefore, the proposed method provides an effective and robust aberration-compensation strategy that corrects system aberrations across multiple illumination angles in a unified optimization, eliminating the need for separate per-angle correction, and is suited to tasks that require joint optimization across multiple datasets.