Wide-field, real-time limited-angle optical diffraction tomography using massively parallel data processing

We present a GPU-accelerated reconstruction framework enabling real-time imaging in limited-angle optical diffraction tomography (LaODT). The implementation leverages massively parallel computing to achieve fast and efficient volumetric reconstruction while preserving high image quality. Real-time 3D imaging using the direct inversion (DI) method spans 7.3 to 1.26 frames per second for datasets of 350³ to 928³ voxels, respectively. In addition to the real-time DI preview, the library also provides an efficient implementation of the Gerchberg-Papoulis algorithm: 100 iterations require less than 25 s, including host-GPU data transfers, to obtain a high-quality 3D reconstruction from a dataset of 928³ voxels. To facilitate broader adoption of our work within the scientific community, we provide our software as a compiled dynamic-link library (DLL) publicly available on GitHub together with its source code.