Reliable segmentation of anatomical tissues of human head is a major step in
several clinical applications such as brain mapping, surgery planning and
associated computational simulation studies. Segmentation is based on
identifying different anatomical structures through labeling different tissues
through medical imaging modalities. The segmentation of brain structures is
commonly feasible with several remarkable contributions mainly for medical
perspective; however, non-brain tissues are of less interest due to anatomical
complexity and difficulties to be observed using standard medical imaging
protocols. The lack of whole head segmentation methods and unavailability of
large human head segmented datasets limiting the variability studies,
especially in the computational evaluation of electrical brain stimulation
(neuromodulation), human protection from electromagnetic field, and
electroencephalography where non-brain tissues are of great importance.
To fill this gap, this study provides an open-access Segmented Head
Anatomical Reference Models (SHARM) that consists of 196 subjects. These models
are segmented into 15 different tissues; skin, fat, muscle, skull cancellous
bone, skull cortical bone, brain white matter, brain gray matter, cerebellum
white matter, cerebellum gray matter, cerebrospinal fluid, dura, vitreous
humor, lens, mucous tissue and blood vessels. The segmented head models are
generated using open-access IXI MRI dataset through convolutional neural
network structure named ForkNet+. Results indicate a high consistency in
statistical characteristics of different tissue distribution in age scale with
real measurements. SHARM is expected to be a useful benchmark not only for
electromagnetic dosimetry studies but also for different human head
segmentation applications.

通过使用卷积神经网络结构 ForkNet + 对 196 个被试进行了开放获取的分段头部解剖参考模型（SHARM）的生成，在年龄尺度上，结果表明不同组织分布的统计特征与实际测量结果具有高度一致性，SHARM 预计不仅对电磁剂量测定研究，而且对不同的头部分割应用都有重要参考价值。