Long-term repeatability, accuracy, and brain size dependence of automatic VOI localization in single-voxel MRS

Precise localization of the volume of interest (VOI) in SVS is essential for reliable MRS data collection. This study evaluates the repeatability and accuracy of an automatic VOI localization tool (ALLVOI) and examines the potential influence of subject's brain size on these measures.

Repeatability was assessed using the Dice coefficient (DC_rep) in repeated MRS data collected from 51 subjects with an interval of 10-25 months. Accuracy was evaluated using the Dice coefficient (DC_acc) in 127 subjects from two MRS studies. While initial evaluations assumed a fixed VOI size, we also assessed DC without this assumption, allowing the VOI to vary across subjects for optimal coverage of the target region.

DC_rep values ranged from 0.92 to 0.94 across three brain regions. When the fixed VOI size constraint was removed, repeated VOI volumes remained consistent (<0.8 cm³ difference), and had comparable DC_rep values for the same subject. In terms of accuracy, DC_acc values under the fixed-size strategy ranged from 0.75 to 0.83 and exhibited a significant positive correlation with brain volume. However, when a variable VOI size strategy was applied, DC_acc values improved to 0.84-0.88 and the correlation with brain size was no longer evident.

The accuracy of localization exhibited a positive correlation with brain volume for fixed-size VOI strategy; however, this dependence was substantially reduced when employing a variable-size VOI approach.

A variable-size VOI strategy improves anatomical precision, offering advantages for studies where accurate spatial targeting is essential.