X-ray Cherenkov-luminescence tomography reconstruction with a three-component deep learning algorithm: Swin...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• X-ray Cherenkov-luminescence tomography (XCLT) enables molecular excitation localization in tissues using megavoltage (MV) x-ray scanning.
• Standard filtered backprojection (FBP) algorithms for XCLT reconstruction face limitations due to insufficient data from dose constraints, leading to artifacts and reduced image quality.

Purpose of the Study:

• To develop a novel deep learning algorithm for direct reconstruction of emission quantum yield distribution in XCLT.
• To enhance image quality and quantitative accuracy in XCLT compared to conventional methods.

Main Methods:

• A three-component deep learning model integrating a Swin transformer, convolutional neural network (CNN), and a locality module was developed.
• The Swin transformer extracts pixel-level priors from sinograms, the CNN transforms sinogram data to image space, and the locality module enhances feature delivery.
• The model's efficacy was validated using simulations, physical phantoms, and in vivo experiments.

Main Results:

• The deep learning approach demonstrated superior performance over traditional FBP methods, particularly in handling data-limited scenarios.
• Quantitative metrics including mean squared error, peak signal-to-noise ratio (PSNR), and Pearson correlation showed significant improvements compared to FBP.
• The developed Swin-CNN model achieved a 32.1% higher PSNR than the AUTOMAP deep learning method.

Conclusions:

• The proposed three-component deep learning algorithm offers an effective and advanced reconstruction method for XCLT.
• This approach significantly improves XCLT's ability to reconstruct molecular emission distributions with higher fidelity and accuracy.