Abstract
1991
Objectives One of the difficulties of the hybrid MR-PET scanners is the accurate attenuation correction of the PET data acquired with this new technology. The use of segmented T1-weighted images is not applicable due to the difficult separation of bone and air filled cavities. Therefore, new methods for skull extraction must be found before image segmentation is applied. The aim of this study was to examine a new approach for extracting bone in images obtained with the UTE MR sequence.
Methods The UTE sequence provides two images by setting TE1=0.07 ms and TE2=2.46 ms. Three subjects were measured in the Siemens MAGNETOM Trio MR scanner and their data were analyzed. Furthermore, CT images of the same subjects were available. From the UTE images a scatter plot can be generated where the area of the intensities characteristic for bone can be outlined. The basic idea was to define a bone-ROI over this area to separate these intensities from those of all others tissues. Furthermore, one general bone-ROI valid for all datasets was searched. Using the three registered CT images a common ROI contour was found. After applying a brain mask and normalizing to a same maximum the UTE images were segmented. For validation the Dice coefficients (D) between the segmented and the corresponding CT images were calculated. Only those image planes containing brain tissue were considered to determine an average D.
Results The resulting mean D were 0.86±0.11, 0.72±0.14, and 0.69±0.10, respectively. Depending on the head inclination different amounts of the facial structures were included in the evaluated planes leading to less favorable results exhibiting intensity combinations similar to those typical for bone.
Conclusions As indicated by the D values the segmentation method presented here yields an acceptable classification of the skull. It will be optimized using a greater number of subjects