Dual and Multi-Target Cone-Beam X-ray Luminescence Computed Tomography Based on the DeepCB-XLCT Network

Area of Science:

• Medical Imaging
• Biophysics
• Computational Biology

Background:

• Cone-beam X-ray luminescence computed tomography (CB-XLCT) is a hybrid imaging modality using X-ray-excitable nanoparticles.
• CB-XLCT offers greater imaging depth and reduced autofluorescence compared to traditional techniques like bioluminescence tomography (BLT) and fluorescence molecular tomography (FMT).
• The inherent ill-posed nature of the CB-XLCT inverse problem, due to complex excitation and light scattering, limits reconstruction quality.

Purpose of the Study:

• To introduce an advanced deep learning network, DeepCB-XLCT, for high-fidelity CB-XLCT reconstruction.
• To enhance the accuracy and shape restoration of internal nanoparticle distributions in CB-XLCT imaging.
• To overcome limitations of conventional linear models in CB-XLCT reconstruction.

Main Methods:

• Developed an end-to-end three-dimensional deep encoder-decoder network (DeepCB-XLCT).
• Implemented a structural similarity loss (SSIM) for improved target shape fidelity.
• Incorporated a region-specific loss term to focus on areas of interest.

Main Results:

• DeepCB-XLCT demonstrated superior reconstruction accuracy compared to traditional methods in numerical simulations and phantom experiments.
• In vivo experiments showed enhanced contrast-to-noise ratio and shape similarity for two targets.
• Tomographic images with three targets confirmed the potential for multi-target CB-XLCT imaging.

Conclusions:

• The DeepCB-XLCT network effectively improves the quality of CB-XLCT reconstructions.
• The method minimizes reconstruction inaccuracies associated with simplified linear models.
• DeepCB-XLCT shows promise for advanced multi-target molecular imaging applications.