A Task-dependent Investigation on Dose and Texture in CT Image Reconstruction.

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• Minimizing radiation dose in CT scans is critical for reducing stochastic effects.
• Image reconstruction algorithms and tissue texture significantly impact low-dose CT performance.
• Accurate localization and characterization of lung nodules are essential for early cancer detection.

Purpose of the Study:

• To evaluate the impact of different CT dose levels and reconstruction algorithms on lung nodule localization and characterization.
• To investigate the role of tissue texture preservation in low-dose CT imaging.
• To compare task-based performance across various reconstruction methods and radiation doses.

Main Methods:

• Prospective patient study with 133 participants undergoing CT scans at 120kVp and three dose levels (100, 40, 20 mAs).
• Implementation of three reconstruction algorithms: filtered back-projection (FBP), Markov random field with Huber weighting (MRF-H), and texture-preserving MRF (MRF-T).
• Radiologist review of images for nodule identification (target and non-target) and characterization, blinded to dose and algorithm.

Main Results:

• MRF-T at 40mAs achieved characterization performance comparable to FBP at 100mAs.
• MRF-T at 40mAs and 20mAs demonstrated nodule localization performance statistically similar to FBP at 100mAs.
• MRF-H and FBP at lower doses (40, 20mAs) showed statistically significant performance differences compared to FBP at 100mAs.

Conclusions:

• Texture-preserving reconstruction (MRF-T) enhances both nodule localization and characterization at low-dose CT.
• Nodule characterization is more dose-sensitive than localization, requiring either higher radiation doses or advanced reconstruction techniques.
• The study highlights the potential of advanced reconstruction algorithms to maintain diagnostic accuracy at reduced radiation exposure.