Anatomic evaluation of 3-dimensional ultrashort-echo-time bone maps for PET/MR attenuation correction

Ultrashort-echo-time (UTE) sequences have been proposed in the past for MR-based attenuation correction of PET data, because of their ability to image cortical bone. In the present work we assessed the limitations of dual-echo UTE imaging for bone segmentation in head and neck imaging. Sequentially acquired MR and PET/CT clinical data were used for this purpose.

Twenty patients referred for a clinical oncology examination were scanned using a trimodality setup. Among the MR sequences, a dual-echo UTE acquisition of the head was acquired and used to create tissue R2 maps. The different undesired structures present in these maps were identified by an experienced radiologist. Global and local measurements of the overlap between R2-based and CT-based bone masks were computed.

UTE R2 maps displayed a nonfunctional relation with CT data. The obtained bone masks showed acceptable overlap with the corresponding CT data, in the case of the skull itself (e.g., 47% mismatch for the parietal region), with decreased performance in the base of the skull and in the neck (e.g., 78% for the maxillary region). Unwanted structures were detected, both anatomic (e.g., sternocleidomastoid, temporal, and masseter muscles) and artifactual (e.g., dental implants and air-tissue interfaces).

It is indeed possible to estimate the anatomic location of bone tissue using UTE sequences. However, using pure parametric maps for attenuation correction may lead to bias close to certain anatomic structures and areas of high magnetic field inhomogeneity. More sophisticated approaches are necessary to compensate for these effects.