Tomographic Sparse View Selection Using the View Covariance Loss

Area of Science:

• Medical Imaging
• Computational Imaging
• Non-Destructive Evaluation

Background:

• Standard computed tomography (CT) reconstruction algorithms like FBP and FDK require numerous views, increasing acquisition time and cost.
• Developing high-quality CT reconstructions from limited views is crucial for efficient non-destructive evaluation (NDE).
• Optimal view selection for sparse-view CT reconstruction remains an unresolved challenge.

Purpose of the Study:

• To introduce a novel view covariance loss (VCL) function for measuring the joint information content of view subsets.
• To develop fast algorithms for computing VCL and an associated algorithm for selecting optimal views.
• To evaluate the effectiveness of the proposed view covariance loss selection (VCLS) algorithm in sparse-view CT reconstruction.

Main Methods:

• Development of a novel view covariance loss (VCL) function to approximate normalized mean squared error (NMSE).
• Implementation of fast algorithms for VCL computation.
• Design of a greedy algorithm for selecting view subsets that minimize VCL.

Main Results:

• The VCLS algorithm demonstrated superior performance compared to existing methods in simulated and measured data.
• Reconstructions using VCLS exhibited lower normalized root mean squared error (NRMSE).
• VCLS resulted in fewer artifacts and enhanced accuracy in sparse-view CT reconstructions.

Conclusions:

• The proposed VCLS algorithm offers an effective solution for selecting optimal views in sparse-view CT reconstruction.
• This approach significantly improves reconstruction quality and accuracy, addressing a key limitation in NDE applications.
• VCLS provides a valuable tool for reducing scan time and costs in CT imaging while maintaining high image fidelity.