Improving thoracic four-dimensional cone-beam CT reconstruction with anatomical-adaptive image regularization (AAIR)

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computational Anatomy

Background:

• Current thoracic four-dimensional cone-beam computed tomography (4D CBCT) reconstructions using the Feldkamp-Davis-Kress (FDK) algorithm suffer from noise and streaking.
• Total-variation (TV) minimization offers noise reduction but can over-smooth anatomical details and is computationally inefficient.

Purpose of the Study:

• To demonstrate a proof of concept for overcoming the limitations of TV minimization in 4D CBCT.
• To introduce an anatomical-adaptive image regularization (AAIR) method that incorporates thoracic anatomy segmentation into the reconstruction process.

Main Methods:

• The AAIR method utilizes the adaptive-steepest-descent projection-onto-convex-sets (ASD-POCS) framework with an added anatomy segmentation step in each iteration.
• Anatomy segmentation information is heuristically applied to adaptively suppress over-smoothing at critical anatomical structures.
• Method performance was evaluated using a digital phantom and a patient scan, compared against FDK, ASD-POCS, and prior image constrained compressed sensing (PICCS).

Main Results:

• AAIR demonstrated superior accuracy in phantom reconstructions (mean absolute difference, structural similarity index).
• Patient scans showed AAIR achieved the highest signal-to-noise ratio and contrast-to-noise ratios for tumor and bony anatomy.
• AAIR significantly reduced over-smoothing compared to ASD-POCS and avoided artifacts seen in PICCS, with over 50% reduction in computation time compared to ASD-POCS.

Conclusions:

• Incorporating anatomy segmentation into 4D CBCT reconstruction via the AAIR method significantly improves image quality and computational efficiency.
• AAIR effectively balances noise reduction with preservation of anatomical details, outperforming existing methods.
• Further development is needed to enable practical clinical application of AAIR.