Evaluation of substantia nigra volume quantification approaches using neuromelanin-sensitive MRI (NM-MRI)
Abstract
Neuromelanin-sensitive MRI (NM-MRI) provides a non-invasive means of characterizing neuromelanin-rich nuclei in the midbrain. However, accurate delineation of the substantia nigra remains challenging. Here, we evaluated two alternative segmentation approaches--graph-cut (GC) and atlas-based segmentation--to gold-standard manual segmentation (MS) using ultra-high resolution 7T NM-MRI data. The GC approach employed a semi-automated region-growing algorithm guided by user-defined seeds, while the atlas-based method involved warping a previously developed SN atlas into each participant’s native space. We evaluated each method against MS using a comprehensive set of segmentation evaluation metrics. GC segmentation produced SN volumes statistically indistinguishable from MS and demonstrated moderate boundary alignment, suggesting its viability as a labor-saving alternative. In contrast, the atlas-based method substantially overestimated SN volume and showed poor boundary accuracy, underscoring the need for refined preprocessing pipelines. Overall, GC segmentation emerges as a promising tool for reliable, large-scale NM-MRI analyses, reducing reliance on time-intensive manual delineations.