Segment and support: a dual-purpose deep learning solution for limited angle holographic tomography

Holographic tomography (HT) enables volumetric investigation of biological and morphological properties of cells and tissues. In its most popular limited-angle implementation, the calculated tomographic reconstructions of investigated samples suffer from missing cone artifacts. Conventional approaches to this problem proved to be time inefficient and prone to noise and artifacts in the source holograms. In this paper, we propose the Segment and Support (SnSNet) method, which corrects the missing cone problem by a priori generation of volumetric object support, free from the missing cone artifacts. We have shown that such object support allows for the very precise calculation of the samples’ volume. Additionally, it allows for the retrieval of high-quality refractive index distribution while used in an iterative HT reconstruction algorithm as a strong regularization term. SnSNet is an advanced deep-learning-based solution capable of facilitating quantitative analysis and object support generation for biological specimens with HT.