A Monte Carlo based scatter removal method for non-isocentric cone-beam CT acquisitions using a deep convolutional...

Area of Science:

• Medical Imaging Physics
• Radiological Sciences
• Computational Imaging

Background:

• Scattered photons in cone-beam computed tomography (CBCT) degrade image quality by introducing artifacts.
• Existing scatter correction methods may not fully address artifacts in non-isocentric acquisitions.
• The ImagingRing™ system on rails (IRr) enables flexible, non-isocentric imaging geometries.

Purpose of the Study:

• To develop and evaluate a projection-based scatter removal algorithm using a deep convolutional autoencoder (DCAE) for the IRr system.
• To assess the performance of the DCAE for both isocentric and non-isocentric CBCT acquisitions.
• To compare the proposed DCAE scatter correction with existing methods and uncorrected data.

Main Methods:

• A deep convolutional autoencoder (DCAE) model was trained using Monte Carlo simulated projection data from digital head-and-neck cancer patients.
• Simulations included both non-isocentric (DCAE_NONISO) and isocentric (DCAE_ISO) datasets to train and test the scatter removal algorithms.
• The DCAE_NONISO algorithm was applied to real patient CBCT data, and image quality metrics were quantitatively evaluated.

Main Results:

• The DCAE_NONISO demonstrated superior scatter removal performance compared to heuristic methods in phantom and patient simulations.
• DCAE_NONISO corrected images showed significantly improved image quality metrics, including higher contrast-to-noise ratio (CNR) between tissues.
• The algorithm was successfully applied to real non-isocentric patient CBCT acquisitions, achieving statistically significant CNR improvements.

Conclusions:

• A projection-based scatter removal algorithm utilizing a DCAE trained on Monte Carlo data is effective for CBCT imaging.
• The proposed DCAE_NONISO method significantly enhances image quality in both simulated and real non-isocentric CBCT scans.
• This study presents the first successful application of a DCAE-based scatter correction for isocentric and non-isocentric CBCT.