Multiscale local sparsity and prior learning algorithm for Cherenkov-excited luminescence scanned tomography...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Radiation Oncology

Background:

• Cherenkov-excited luminescence scanned tomography (CELST) is an emerging imaging technique with potential in radiation therapy.
• Current CELST methods suffer from low resolution and image quality degradation due to light scattering and limited data.
• Inaccurate probe distribution information hinders effective application in therapy.

Purpose of the Study:

• To develop a novel reconstruction method for CELST to improve image quality and accurately characterize luminescent probes.
• To address the limitations of low resolution and degraded image quality in existing CELST techniques.
• To enhance the quantitative accuracy of luminescent probe imaging for radiation therapy applications.

Main Methods:

• Proposed a novel reconstruction method, LKSVD-Net, combining a sparse prior with an attention network for CELST.
• Incorporated a multiscale learned KSVD to capture local sparsity information of luminescent probes.
• Designed a prior attention network to leverage measurement-related prior features and combined them for image reconstruction.

Main Results:

• LKSVD-Net significantly enhances image quality, even at a 20 dB signal-to-noise ratio (SNR).
• The method achieves improved quantitative accuracy for small probes with close proximity.
• LKSVD-Net demonstrated substantial improvements in PSNR (~15.1%), SSIM (~95.8%), and PC (~3%) over Tikhonov regularization.

Conclusions:

• LKSVD-Net offers a significant advancement in CELST image reconstruction, overcoming current resolution and quality limitations.
• The proposed method provides accurate characterization of luminescent probes, vital for advancing radiation therapy.
• LKSVD-Net shows superior performance compared to traditional methods, paving the way for improved clinical applications.