Estimate and compensate head motion in non-contrast head CT scans using partial angle reconstruction and deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computed Tomography

Background:

• Head motion during CT scans causes artifacts, degrading image quality and potentially leading to misdiagnoses.
• Partial Angle Reconstruction (PAR) has shown promise in cardiac CT but its application in head CT for motion artifact reduction is unexplored.

Purpose of the Study:

• To develop a deep learning (DL) model for direct head motion estimation from PAR head CT images.
• To integrate DL-based motion estimation into iterative reconstruction for artifact compensation.

Main Methods:

• A convolutional neural network (CNN) was trained to estimate B-spline control points representing head motion (translation and rotation) from 25 consecutive PAR segments.
• Estimated motion parameters were incorporated into an iterative reconstruction algorithm's forward and backprojection steps.
• Model performance was validated using both simulated data and physical phantom studies.

Main Results:

• The DL model accurately estimated head motion with low mean absolute error (MAE) in both simulation (0.28-0.45 mm/degree) and phantom studies (0.40-0.48 mm/degree).
• Motion-corrected CT images showed substantial artifact reduction (simulation MAE: 178 to 37 HU; phantom MAE: 117 to 42 HU).
• Image quality significantly improved, with Structural Similarity Index (SSIM) increasing from 0.60 to 0.98 (simulation) and 0.83 to 0.98 (phantom).

Conclusions:

• The proposed DL model effectively estimates head motion using PAR data.
• This approach significantly reduces motion artifacts in head CT, improving diagnostic accuracy.