The neural network accelerates the reconstruction of the 3D image of biological samples

Comparison of internal and external circularization of FIN and RH-M on lung tissue sections, salivary glands, and Pap smear samples. The results of the MH-PR reconstruction using the same input holograms (M = 3) are also shown for comparison. Credit: Hanlong Chen, UCLA

Researchers have developed a new comprehensive neural network that can speed up reconstruction of 3D images. Unlike other deep learning techniques, this approach can be used on samples that were not encountered during training, making it particularly useful for label-free 3D biomedical imaging.

“With this frame, well trained neural network They can be distributed anywhere, without fine-tuning, and perform fast, high-quality 3D imaging of different samples,” explained research leader Hanlong Chen, University of California, Los Angeles (UCLA).

Hanlong Chen and Aydogan Ozkan will present the research at the Frontiers Conference in Optics + Laser Sciences (FiO LSThe meeting took place in Rochester, NY and online from October 17-20, 2022. Submission is scheduled for Monday, October 17 at 16:30 EST (UTC-04:00).

generalizable approach

Although many neural networks have been developed to achieve the data-heavy task of hologram reconstruction biological research And the Biomedical applicationsMost are designed to be very specific. This means that it may not perform well if used with samples other than those initially used for training network.

To solve this problem, Chen and his colleagues developed a comprehensive neural network called the Fourier Imager Network (FIN). This type of neural network is trained using a single model, bypassing some steps typically used by other deep learning methods. End-to-end neural networks are also faster and more generalizable to a variety of samples.

Faster and more accurate results

The FIN framework takes a series of raw-only density holograms taken at different distances from sample to sensor using an embedded lens-free 3D microscope and creates reconstructed images of the samples. To test the new approach, the researchers trained the network using sections of lung tissue. They then used FIN to reconstruct 3D images of human salivary gland tissue and Pap smear samples that the network had not seen during training.

FIN worked well on these new types of samples and provided images that were more accurately reconstructed than an iterative algorithm and a modern deep learning model. It also showed an improved speed of nearly 50 times compared to the deep learning model. The researchers say that these results demonstrate the strong extrinsic generalization of FIN while also demonstrating the enormous potential of deep constructing a large-scale generalizable. neural networks For various microscopy and computational imaging tasks.

Chen added, “Our next step is to investigate AF while retaining the advantages of our approach, such as superb image quality, unprecedented generalization of new sample types, and improved computational speed, making holographic imaging possible with low-resource hardware.”


Super phase recovery and hologram reconstruction using a deep neural network


the quote: Neural Network Speed ​​for 3D Image Reconstruction of Biological Samples (2022, September 22) Retrieved on September 22, 2022 from

This document is subject to copyright. Notwithstanding any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.