MRS Metabolite Basis Set & LCModel Spectral Fitter

Magnetic Resonance Spectroscopy · Chemical Shift Dispersion, Basis Sets, Baseline Deconvolution & Cramer-Rao Lower Bounds (CRLB)

Clinical Presets:
Synthetic Data
Total Fit
Baseline
Residual (Offset)
Metabolite Resonances (δ ppm) True Rel. Conc Fit Conc CRLB (%SD) Reliability
NAA / Cr Ratio
--
Norm: 1.5 – 2.1
Cho / Cr Ratio
--
Norm: 0.7 – 0.9
mI / Cr Ratio
--
Norm: 0.6 – 0.8
Main Magnetic Field (B0) 3.0 T (127.7 MHz)
Shim Quality (Linewidth FWHM) 5.0 Hz
Excellent Shim (2 Hz) Poor Shim (18 Hz)
Signal-to-Noise Ratio (SNR) 25
Phase Offset (φ0) 0°
Macromolecule Baseline Medium
Overlay Basis Components:

Physical Principles & Mathematical Formulation

1. Chemical Shift Dispersion & Field Strength:

Electrons orbiting distinct chemical moieties generate local diamagnetic shielding currents that oppose the main field: \(B_{\text{local}} = B_0(1 - \sigma)\). Chemical shifts are reported on a dimensionless parts-per-million scale (\(\delta = (\nu - \nu_{\text{ref}})/\nu_{\text{ref}} \times 10^6\)). While the chemical shift in ppm is field-independent, frequency dispersion in Hertz increases linearly with field strength (\(\Delta\nu_{\text{Hz}} = \gamma B_0 \Delta\delta\)). At 7.0 T, J-coupled multiplets (Glutamate, GABA) and crowded singlets (Cho, Cr) separate cleanly, reducing spectral covariance and fitting errors.

2. Basis Sets vs. Single-Peak Fitting:

In-vivo 1H spectra consist of overlapping scalar J-coupled multiplets and a broad macromolecular envelope. Rather than fitting unconstrained individual Gaussians, the Linear Combination Model constructs the total signal from quantum-mechanically simulated or phantom-recorded metabolite basis spectra:

\[ S(\delta) = \sum_k c_k B_k(\delta) \otimes L(\Delta\nu) + \text{Baseline}(\delta) \]
This fixes relative peak multiplet areas, chemical shifts, and J-splitting within each molecule, leaving only overall concentration \(c_k\), line broadening, and baseline to be estimated.

3. Cramer-Rao Lower Bounds (CRLB):

The Cramer-Rao inequality establishes the minimum possible variance achievable by any unbiased estimator:

\[ \text{Cov}(\hat{c}) \ge F^{-1} = \sigma^2 (X^T X)^{-1} \]
where \(X\) contains the metabolite basis columns and \(\sigma^2\) is the noise variance. The relative uncertainty \(\text{CRLB} = (\text{SD}_k / \hat{c}_k) \times 100\%\) reflects not only noise, but severe covariance between overlapping metabolites. By convention in clinical spectroscopy, metabolite fits with \(\text{CRLB} > 20\%\) are classified as unreliable.

4. Diagnostic Neurochemical Markers:
  • NAA (2.02 ppm): Synthesized exclusively in neuronal mitochondria; marker of viable neuronal density. Drops during neurodegeneration, stroke, and neoplastic infiltration.
  • Creatine (3.03 ppm): Phosphocreatine energy reservoir; often serves as an internal concentration reference denominator.
  • Choline (3.22 ppm): Phosphocholine and glycerophosphocholine; elevated during cellular membrane synthesis and turnover (hallmark of active gliomas).
  • Lactate (1.33 ppm): Anaerobic glycolysis end-product; appears as an inverted/upright doublet in acute ischemia, mitochondrial disorders, and necrotic cysts.