N4 Bias Filter

N4 Bias Filter applies the N4ITK bias field correction to magnetic resonance (MR) images to mitigate spatial intensity inhomogeneity and improve MR-based quantification.

Key Features:

• Bias Field Correction: Applies the N4 filter to T2-weighted MR images using either default parameters or user-defined settings to reduce bias field effects.
• Parameter Optimization: Systematically explores convergence threshold, shrink factor, fitting level, number of iterations, and mask usage to identify optimal N4 configurations.
• Optimization Metric: Uses the Full Width at Half Maximum (FWHM) of the segmented periprostatic fat distribution as the evaluation metric for parameter selection.
• Quantitative Evaluation: Assessed the performance of 240 different N4 configurations using phantom T2-weighted images to simulate uniform tissue conditions.
• Validation Across Datasets: Validated selected configurations on two public datasets of 89 and 204 T2-weighted patient images acquired with different coils at 1.5 T and 3 T, plus two external datasets.
• Recommended Configurations: Reports optimal parameter sets including (a) combined surface and endorectal coil at 1.5 T and 3 T: convergence threshold 0.001, shrink factor 2, fitting level 6, iterations 100, default mask; and (b) surface coil at 1.5 T or 3 T: convergence threshold 0.001, shrink factor 2, fitting level 5, iterations 25, default mask.

Scientific Applications:

• Pelvic prostate imaging: Produces bias-corrected T2-weighted images to support accurate MR-based quantification in prostate studies.
• Segmentation and characterization: Enables robust input for segmentation and characterization of periprostatic fat distribution using the FWHM metric.
• Clinical and research quantification: Improves reliability of quantitative measurements used in diagnosis and treatment planning for prostate imaging.

Methodology:

Apply the N4ITK filter to MR images, perform a systematic parameter sweep across convergence threshold, shrink factor, fitting level, number of iterations, and mask usage, evaluate configurations using the FWHM of segmented periprostatic fat distribution (240 configurations tested on phantom T2-weighted images), and validate selected configurations on public and external T2-weighted patient datasets acquired with surface and endorectal coils at 1.5 T and 3 T.