A Post-processing Technique to Suppress Fluid Signal and Increase Contrast in Multispectral MR Exams of MS Patients

Joseph Ross Mitchell, Paula J. Gareau Foster, Stephen J. Karlik, Brian K. RuttMICCAI 1999 · DOI

Abstract

Multispectral Magnetic Resonance Imaging (MRI) provides a wealth of diagnostic information but often complicates clinical interpretation due to high dimensionality and varying tissue contrasts. This study introduces a "spectral phase" (SP) imaging technique that collapses 4D multispectral data into a single, high-contrast image. By utilizing cerebrospinal fluid (CSF) as a reference vector in feature space, the method retrospectively suppresses fluid signals and enhances tissue boundaries. Quantitative analysis demonstrates contrast improvements of up to 187% for CSF vs. white matter, providing a clearer "anatomical photograph" of the brain and revealing subtle heterogeneity in Multiple Sclerosis (MS) lesions.

Introduction

Magnetic Resonance Imaging (MRI) is the gold standard for detecting Multiple Sclerosis (MS) lesions. Clinicians typically rely on multispectral exams, which include T1-weighted (T1w), T2-weighted (T2w), and Proton Density-weighted (PDw) scans. While these multiple sequences provide complementary information, they also introduce a "curse of dimensionality." As the number of images per exam increases, the mathematical "feature space"—where each pixel's intensity across different scans is plotted—becomes too complex for efficient manual analysis. Furthermore, standard scans often suffer from intensity inhomogeneities and poor contrast between specific tissue types. This paper, originally presented at MICCAI 1999, proposes a geometric post-processing method to extract the most relevant diagnostic information into a single, intuitive image.

Methodology

The proposed approach relies on a geometric interpretation of MRI data within a multidimensional feature space. The workflow consists of four primary stages:

  1. Multispectral Acquisition: Four distinct MR contrasts (PDw, T2w, T1w, and T1w with Gadolinium) were collected for each patient using a 1.5T scanner.
  2. Preprocessing: To ensure data quality, non-parametric intensity normalization was applied to correct "shading" artifacts. Additionally, a non-linear anisotropic diffusion filter was used to reduce noise while preserving critical edge information.
  3. Spectral Phase (SP) Calculation: The core novelty involves defining a 4D "reference vector" based on the average intensity of cerebrospinal fluid (CSF). For every voxel in the brain, the "spectral phase" is calculated relative to this CSF vector. Tissues spectrally similar to CSF result in low values (suppression), while tissues that are spectrally distinct (like white or gray matter) result in high values.
  4. Quantitative Evaluation: Contrast was measured using a Signal Difference to Noise Ratio (dSNR) formula to compare the new SP images against traditional MRI sequences.

Results and Discussion

The quantitative gains in contrast were substantial across all tissue comparisons. The CSF vs. White Matter (WM) contrast increased by 187% (from a dSNR of 19.5 in T1w to 56 in SP), while CSF vs. Gray Matter (GM) contrast increased by 143%. Even the subtle distinction between WM and GM saw a 53% improvement in contrast compared to standard T2w scans.

Comparison of standard MRI sequences (a: PDw, b: T2w, c: T1w) with the processed CSF-SP image (d), demonstrating superior contrast and fluid suppression.

Beyond quantitative metrics, the SP images provide an "anatomical photograph" effect, making them highly intuitive for clinicians. A key clinical insight was the visualization of lesion heterogeneity; where standard T1 or T2 scans might show a uniform "blob," the SP images revealed internal structures and subtle changes in surrounding tissue. In a longitudinal case study, the SP technique successfully identified a diffuse abnormality that appeared and receded over five months, a change that was significantly harder to quantify in standard sequences.

Conclusion

The spectral phase technique offers a principled way to reduce 4D multispectral data into a 1D image while simultaneously increasing the useful signal. By allowing for retrospective fluid suppression, it provides the benefits of sequences like FLAIR without requiring them to be decided upon during the initial scan. While further validation through larger cohorts and ROC analysis is required, this method represents a significant step toward simplifying the analysis of complex MS imaging data.

Summary updated

BibTeX
@inproceedings{mitchell1999post,
  author = {Joseph Ross Mitchell and Paula J. Gareau Foster and Stephen J. Karlik and Brian K. Rutt},
  title = {{A Post-processing Technique to Suppress Fluid Signal and Increase Contrast in Multispectral MR Exams of MS Patients}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {1999},
  doi = {10.1007/10704282\_24},
}
Figure