Deep learning-based 3D refractive index generation for live blood cell

We propose a novel method to represent the inside and outside structures of a living biological sample with a label-free process using deep learning. The proposed approach combines the 3D refractive index characteristics with the deep learning method, making a new contribution to the bioimaging fields. In particular, the proposed deep learning model produces numerical 3D refractive indexes, which can not only provide important biometric information for the medical field but also analyze the statistical elements directly from the numerical output. We acquired a data set consisting of multi-viewed holographic images and 3D refractive index images of living blood-cell samples through holographic tomogram microscopy. We found that the proposed model’s PSNR is 10.17dB and 3D visualization of the generated 3D refractive index data is reasonably performed by comparing it with the ground truth.