Residual Water Suppression in MRS Using HLSVD With Automatic Component Selection Strategy

ABSTRACT Residual water signals in magnetic resonance spectroscopy (MRS) can distort baselines and complicate metabolite quantification, particularly in regions with poor field homogeneity and in MRSI. The Hankel Lanczos singular value decomposition (HLSVD) method is widely used for Residual water removal, but its performance depends on the number of components selected. This study proposes an automatic approach that combines HLSVD with the residual water to metabolite variance ratio (RWVR) for component selection. The RWVR, defined as the variance of spectral fluctuations in the water region (4.2–5.2 ppm) normalized to the metabolite region (0.5–4.0 ppm). HLSVD was applied using component numbers between 10 and 32, and the configuration yielding the minimal RWVR was chosen as optimal. The method was evaluated on single voxel spectroscopy (SVS) data from four brain regions and on MRSI datasets acquired at 3 T. The Residual Water Index, defined as the variance ratio before and after RW removal, was used to assess suppression performance. Optimal component numbers were more frequently found in the range of 22–32, consistent with prior reports. With the identified component numbers, acceptable spectra across both SVS and MRSI can be produced. While multiple component numbers often yielded similar outcomes, some choices led to unacceptable spectra with baseline distortions. RWVR provided an effective indicator for avoiding such outcomes. Residual water index values were generally low, confirming effective suppression, though a small subset of cases exhibited irregular residual water lineshapes and higher RWVR. By evaluating multiple component numbers and choosing the configuration that minimizes RWVR, the proposed method avoids manual tuning and ensures acceptable suppression across both SVS and MRSI data. This approach offers a practical solution for large‐scale studies where consistent and reliable residual water removal is essential.