Fast compressed sensing-based CBCT reconstruction using Barzilai-Borwein formulation for application to on-line IGRT

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• Compressed sensing enables accurate, low-dose cone-beam computed tomography (CBCT) reconstruction.
• Reconstruction time is a major limitation for clinical CBCT implementation.

Purpose of the Study:

• To develop a novel, efficient gradient projection algorithm for CBCT reconstruction.
• To address the challenge of long reconstruction times in low-dose CBCT.

Main Methods:

• A Gradient-Projection-Barzilai-Borwein (GP-BB) algorithm was developed for total variation (TV)-norm regularization-based least squares CBCT reconstruction.
• Multiresolution optimization and single Graphics Processing Unit (GPU) implementation were used to accelerate the process.
• The algorithm was evaluated using numerical phantoms, a physical phantom, and clinical head-and-neck patient data, comparing it against other compressed sensing methods and conventional FDK reconstruction.

Main Results:

• The GP-BB algorithm demonstrated faster convergence (≤30 iterations) compared to other compressed sensing algorithms (≥50 iterations).
• Clinically acceptable images were reconstructed from 40 projections in under 12.6 seconds using the CatPhan phantom.
• For head-and-neck scans, 120 projections yielded images comparable or superior to conventional FDK reconstructions using 364 projections, achieving a ~67% dose reduction.

Conclusions:

• A novel, fast, low-dose CBCT reconstruction algorithm (GP-BB) was successfully developed.
• The algorithm achieves clinically viable image quality with significant dose reduction (~67%) and acceptable reconstruction times (34-78 seconds).
• The GP-BB algorithm shows promise for on-line image-guided radiation therapy (IGRT).