Sparse-view CT reconstruction based on multi-level wavelet convolution neural network.

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Sparse-view computed tomography (CT) reduces radiation dose and acquisition time.
• Iterative algorithms (e.g., total-variation) and U-NET deep learning show promise but face challenges like computational burden and image artifacts.
• Developing advanced reconstruction methods is crucial for sparse-view CT efficacy.

Purpose of the Study:

• To propose and evaluate a novel multi-level wavelet convolutional neural network (MWCNN) for sparse-view CT reconstruction.
• To address limitations of existing methods, including artifact reduction and computational efficiency.
• To enhance image quality in sparse-view CT by preserving anatomical structures.

Main Methods:

• A multi-level wavelet convolutional neural network (MWCNN) architecture was developed, integrating wavelet transform with a modified U-NET (without pooling).
• Filtered backprojection (FBP) was used for initial reconstruction of sinograms from 60, 120, and 180 projections.
• The MWCNN processed FBP-reconstructed sparse-view data, utilizing wavelet transform to enlarge the receptive field and improve performance.

Main Results:

• The proposed MWCNN method demonstrated superior performance in reducing streaking artifacts and preserving anatomical structures compared to interpolation, iterative TV, and standard U-NET methods.
• Quantitative evaluations using structural similarity, root mean square error, and resolution metrics confirmed the MWCNN's effectiveness.
• The MWCNN achieved the highest performance across various evaluation parameters.

Conclusions:

• The multi-level wavelet convolutional neural network (MWCNN) shows significant potential for high-quality sparse-view CT reconstruction.
• This deep learning approach offers an effective solution for overcoming challenges associated with sparse-view CT imaging.
• The MWCNN method represents a promising advancement in medical imaging reconstruction techniques.