Feature constrained compressed sensing CT image reconstruction from incomplete data via robust principal component...

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computer-Aided Diagnosis

Background:

• Computed tomography (CT) image reconstruction suffers from artifacts due to incomplete data, particularly limited angle projections.
• Existing methods like prior image constrained compressed sensing require pre-scans of the same patient, which are not always available.
• Clinical databases offer a rich source of well-reconstructed images from different patients that can serve as prior knowledge.

Purpose of the Study:

• To propose a novel feature constrained compressed sensing (FCCS) algorithm for CT image reconstruction.
• To improve image quality in limited angle CT reconstruction by leveraging prior knowledge from a clinical image database.
• To address the challenge of artifact reduction when a pre-scan of the same patient is unavailable.

Main Methods:

• Developed a feature constrained compressed sensing (FCCS) algorithm for CT image reconstruction.
• Employed robust principal component analysis (RPCA) to extract features from a clinical image database, creating a low-dimensional linear space.
• Formulated a bi-criterion convex program combining feature constraints and total variation constraints for the reconstruction procedure.

Main Results:

• The FCCS algorithm successfully improved image quality in CT reconstruction, particularly for limited angle problems.
• Numerical simulations using both phantom and real clinical patient images demonstrated the effectiveness of the proposed algorithm.
• The method effectively utilized prior knowledge from dissimilar but related images to enhance reconstruction quality.

Conclusions:

• The proposed FCCS algorithm offers a promising solution for CT image reconstruction with incomplete data, especially limited angle projections.
• Leveraging prior knowledge from clinical databases via feature extraction provides a viable alternative when same-patient prior scans are unavailable.
• The algorithm demonstrates significant potential for reducing artifacts and enhancing diagnostic accuracy in clinical CT imaging.