X-ray cone-beam computed tomography geometric artefact reduction based on a data-driven strategy.

Area of Science:

• Medical Imaging
• Computer Vision
• Materials Science

Background:

• Cone-beam computed tomography (CBCT) provides high-accuracy 3D imaging but is susceptible to geometric artifacts from system misalignment.
• These artifacts degrade image quality, causing detail loss and reduced spatial resolution, hindering defect detection in critical industries like aerospace.
• Accurate defect identification is crucial for quality control in precision manufacturing and material analysis.

Purpose of the Study:

• To develop and validate a novel method for reducing geometric artifacts in CBCT images.
• To improve the accuracy of defect detection in industrial applications by enhancing image quality.
• To present a data-driven strategy using a modified fully convolutional neural network (M-FCNN) for artifact reduction.

Main Methods:

• An end-to-end modified fully convolutional neural network (M-FCNN) was designed, featuring five convolution and five deconvolution layers.
• The M-FCNN architecture omits pooling layers to preserve image details, crucial for high-resolution reconstruction.
• The network was trained separately on artifact images with diverse features and validated on synthetic and practical datasets.

Main Results:

• The M-FCNN effectively reduced geometric artifacts in CBCT images across various synthetic data types.
• Validation with practical data, including carbon composite and medical oral phantoms, demonstrated comparable or superior quality to existing methods.
• The method proved successful in enhancing image quality in the image domain through a data-driven strategy.

Conclusions:

• The proposed M-FCNN offers an effective solution for mitigating geometric artifacts in CBCT imaging.
• This data-driven approach significantly improves image quality, supporting more accurate defect detection in industrial settings.
• The M-FCNN presents a promising tool for enhancing the reliability and precision of CBCT applications.