A hybrid reconstruction algorithm for fast and accurate 4D cone-beam CT imaging

Area of Science:

• Medical Imaging
• Radiation Oncology
• Image Reconstruction

Background:

• 4D cone beam CT (4D-CBCT) is crucial for image guidance in radiation therapy, particularly for lung and upper abdomen treatments.
• Current clinical use of 4D-CBCT is hindered by long scan times and suboptimal image quality.

Purpose of the Study:

• To develop an advanced 4D-CBCT reconstruction technique.
• To restore high-quality volumetric images from data acquired during a standard 1-minute 3D-CBCT scan.

Main Methods:

• A novel algorithm optimizes a deformation vector field to align patient-specific planning CT (p-CT) with 4D-CBCT projections.
• A forward-backward splitting (FBS) method iteratively solves subproblems of image reconstruction and deformable image registration.
• The workflow is GPU-accelerated for enhanced efficiency and evaluated on phantoms and patient cases.

Main Results:

• The algorithm successfully reconstructs 4D-CBCT images from highly under-sampled 1-minute scan data.
• Excellent anatomical accuracy was achieved, with average differences of 0.204 mm and maximum differences of 0.3-0.5 mm.
• Image quality showed intensity errors below 5 HU (phantom) and 20 HU (patients), with significant signal-noise-ratio improvements (12.74x for 1-min data).

Conclusions:

• High-quality 4D-CBCT imaging is feasible using a standard 1-minute 3D-CBCT protocol.
• The proposed hybrid reconstruction algorithm overcomes limitations of current 4D-CBCT techniques.
• This advancement supports improved 4D image guidance in clinical radiation therapy.