SU-F-I-23: 1H Magnetic Resonance Spectroscopy Baseline Correction WithSingular Value Decomposition Method




Baseline distortions in NMR spectra are caused by the corruption of the first few data points in Free Induction Decay (FID) or originated from macromolecules and lipids. To study the Singular Value Decomposition (SVD) based baseline correction method.


In-house Singular Value Decomposition (SVD) program was developed with Matlab version 7.6 ( Hankel SVD method (singular value decomposition of the acquired FID signal arranged in a Hankel matrix) is used to compute the signal poles and amplitude, and from them the signal frequencies and damping factors. The estimated FID signals are constructed from the quantification parameters; residue error spectrum was calculated by FFT of the difference of original FID and estimated FID. In order to study the baseline correction effect, known simulated FID with added noise to mimic known metabolites concentration was analyzed with SVD, quantification results were compared with known reference; then baseline distortion was introduced by removing the first 5 data points, after baseline corrected with SVD method, the quantification results were compared with original reference without baseline distortion.Known patient FID acquired with Siemens Verio 3T scanner was introduced baseline distortion by removing the first 3 data points, after baseline corrected with SVD method, corrected FID were compared with original results without baseline distortion.


Quantification results for simulated data were compared with reference, all within 10% deviation with SNR 20.For FID acquired with Siemens Verio 3T scanner, distorted FID and calculated baseline with SVD method was compared, the baseline matches the distortion well. Corrected FID were used for quantification, the quantification results match the corrected FID well.


As demonstrated, SVD based baseline correction method can be used to correct the FID data point missing type of distortion, results are satisfactory for both simulated data and acquired patient data.