Patch-based artifact reduction for three-dimensional volume projection data of sparse-view micro-computed tomography.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence

Background:

• Micro-computed tomography (micro-CT) provides non-destructive 3D morphological analysis at the micrometer scale.
• Current micro-CT imaging of tissues is limited by long scan times, hindering its use in pathology and histology.
• Sparse-view CT reduces scan time and radiation dose but introduces severe streak artifacts due to undersampling.

Purpose of the Study:

• To develop an effective artifact reduction method for 3D volume projection data from sparse-view micro-CT.
• To improve the quality of reconstructed tomographic images for better 3D microstructure analysis.
• To enable faster and more efficient micro-CT imaging of biological tissues.

Main Methods:

• Development of a patch-based lightweight fully convolutional network (FCN).
• The FCN estimates full-view 3D volume projection data from sparse-view data.
• Evaluation using physically acquired micro-CT datasets.

Main Results:

• The proposed method significantly suppressed streak artifacts in reconstructed images.
• High estimation accuracy was achieved for the full-view data.
• Both training and prediction times for the network were found to be short.

Conclusions:

• The developed FCN is a potent method for artifact reduction in sparse-view micro-CT data.
• This technique has great potential for enhancing 3D tissue microstructure analysis in pathology and histology.
• The method offers a promising solution for faster, high-quality micro-CT imaging.