Low-dose x-ray tomography through a deep convolutional neural network.

Area of Science:

• * Materials Science and Biological Imaging
• * Advanced X-ray imaging techniques

Background:

• * Synchrotron X-ray tomography enables high-resolution imaging of materials and biological tissues.
• * Radiation-sensitive samples face a trade-off between acquisition time, sample damage, and signal quality.
• * Existing low-dose imaging methods often compromise signal-to-noise ratio and structural detail.

Purpose of the Study:

• * To develop a deep convolutional neural network (CNN) method for enhancing low-dose X-ray tomographic projections.
• * To improve signal quality in fast X-ray acquisitions of radiation-sensitive samples.
• * To enable high-resolution imaging with reduced radiation exposure.

Main Methods:

• * Development and application of a deep convolutional neural network (CNN) to enhance X-ray projection quality.
• * Utilizing short-exposure-time projections with CNN-based signal enhancement.
• * Validation with simulated data and experimental transmission X-ray microscopy of mouse brains.

Main Results:

• * CNN-enhanced projections achieved signal-to-noise ratios comparable to long-exposure-time projections.
• * The method increased the signal in low-dose, fast acquisitions by at least a factor of 10.
• * Enhanced datasets allowed for reliable automated tracing of brain structures, outperforming other post-processing techniques.

Conclusions:

• * The CNN method significantly improves the quality of low-dose X-ray tomographic projections.
• * This approach mitigates the trade-off between acquisition speed, radiation dose, and image quality.
• * The technique is applicable to various X-ray imaging modalities and holds potential for studying dynamic processes.