Lag-Net: Lag correction for cone-beam CT via a convolutional neural network

Area of Science:

• Medical Imaging
• Computed Tomography
• Image Processing

Background:

• Charge traps in amorphous silicon detectors cause lag signals, leading to ghosting and artifacts in CT imaging.
• Traditional Linear Time-Invariant (LTI) correction methods are often incomplete and fail to account for exposure dependency, leaving residual artifacts.

Purpose of the Study:

• To develop and evaluate a novel deep learning method (Lag-Net) for accurate lag signal measurement and artifact suppression in CT.
• To overcome the complexity and high instrumentation demands of existing hardware correction methods.

Main Methods:

• A novel hardware correction method was developed, requiring two scans with adjusted CT instrumentation timing to measure lag signals.
• A deep learning method, Lag-Net, was introduced to remove lag signals, using hardware-corrected, nearly lag-free images as training data.

Main Results:

• Lag-Net significantly outperformed traditional LTI correction in suppressing lag artifacts and enhancing image quality on both simulated and real datasets.
• The deep learning method achieved reconstruction results comparable to hardware correction but without its operational complexities.

Conclusions:

• The hardware correction method shows superior artifact suppression, especially at low exposures, despite its complexity.
• Lag-Net effectively bypasses hardware correction's operational drawbacks using deep learning, outperforming LTI correction in low-exposure scenarios.