Modeling subcutaneous fat improves skull segmentation for individualized volume conductor head models

Accurate tissue segmentations are the basis for volume conductor models used for simulating electric field propagation in Transcranial Brain Stimulation (TBS) and for source localization in magneto- and electroencephalography (M/EEG). CHARM is a pipeline for head model creation that uses a parametric Bayesian model for tissue segmentation and is optimized to segment head and brain structures from a fat-suppressed T1-weighted magnetic resonance (MR) scan in combination with a T2-weighted MR scan. However, commonly available datasets often lack this specific combination of scans, leading to suboptimal segmentation results, particularly of the scalp and skull. Bone segmentation from MR scans is generally challenging, due to the low MR signal of compact bone, and is further complicated by conventionally used structural scan sequences, which are often optimized to maximize brain tissue contrast. Here we present an updated version of CHARM with an extended tissue probability atlas, which includes the sub-cutaneous fat, along with optimized segmentation parameters. The updated configuration improves the segmentation accuracy of the skull on all tested input MR sequence combinations compared to the current standard configuration used in CHARM.