Original Paper
Segmentation and visualization of the human cranial bone by T2* approximation using ultra-short echo time (UTE) magnetic resonance imagingSegmentierung und Visualisierung des menschlichen Schädelknochens durch T2* Approximierung mittels Magnetresonanztomographie ultra-kurzer Echozeiten (UTE)

https://doi.org/10.1016/j.zemedi.2019.06.003Get rights and content

Abstract

Segmentation of the human cranial bone from MRI data is challenging, because compact bone is characterized by very short transverse relaxation times and typically produces no signal when using conventional magnetic resonance imaging (MRI) sequences. In this work, we propose a fully automated segmentation algorithm, which uses dual-echo, ultra-short echo-time (UTE) MRI data. The segmentation was initialized by interval thresholding of approximated T2* relaxation time maps in the range of 1 ms < T2* < 3 ms. This parameter range was derived from a manual region-of-interest analysis of high resolution data of the cranial layers, resulting in average T2* relaxation times of 1.7 ± 0.3 ms in the lamina externa, 2.5 ± 0.3 ms in the diploe and 1.7 ± 0.2 ms in the lamina interna. Segmentations were performed based on data of 8 healthy volunteers that were acquired with different acquisition parameters and spatial resolutions to test the stability of the algorithm. Comparison with computed tomography data demonstrated close agreement with the segmented UTE MRI data. Visualization of the segmented cranial bone was performed by volumetric rendering and by using the approximated T2* values for color coding, clearly visualizing the cranial sutures as well as their intersections.

Introduction

Skull segmentation in medical images is an important step towards complete tissue segmentation in the human head, but includes also other aspects of cranial morphology analysis that benefit from it. Examples include image-based information to cranioplasty, assessment of abnormal bone growth or morphological changes with age [1], or the construction of realistic models of the head [2]. Inverse problems occurring in magnetoencephalography (MEG) and electroencephalography (EEG), for instance, require realistic models of the head to accurately compute the mapping from neural current sources to scalp potentials and extra-cranial magnetic fields [3]. Due to different conductivities of the skull and soft tissue, it is crucial that bone regions are included correctly and on an individual basis in such head models.

Compact bone is characterized by very short T2* relaxation times and relatively short T1 relaxation times. Although exact values are not available for the skull, studies investigating compact cortical bone in the lower leg reported T1 relaxation times of 80 ms [4] and 223 ms [5] as well as T2* relaxation times as short as 0.4 ms [5] at 3 T. By using ultra-short echo-time (UTE) imaging with echo-times less than 1 ms, it is, however, possible to directly image compact bone despite the quickly decaying signals [6], [7], [8]. Recent applications of UTE imaging of the head include attenuation correction for combined MR/PET scanning [9], [10], [11], [12], [13], skull aberration correction in MR-guided transcranial focused ultrasound [14], or synthetic CT data generation in radiation therapy planning [15]. Although UTE sequences make it possible to visualize compact bone structures, segmentation of bone and other short T2* tissues from tissues with long T2* components represent a major challenge. Since UTE data are collected with very short echo-times and short repetition times, typically heavily T1-weighted images are obtained. Because of the short time between excitation and data sampling, the development of tissue contrast based on T2/T2* dephasing is low, and UTE-based images therefore usually show very little contrast between tissues. Several methods have been proposed to overcome the challenge of separating or segmenting short T2* tissues from their surroundings. The simplest approach is to apply a threshold to multi-echo UTE images after subtracting a late echo image, where the short T2* components have mostly decayed, from the first early echo image [16], [17]. Alternatively, thresholded T2* maps estimated from dual-echo UTE by linear approximation data can be used for segmentation [9]. Another approach, using zero echo-time (ZTE) imaging with the shortest possible echo-time “close to zero”, also demonstrated convincing segmentation results, based on thresholded, logarithmically scaled ZTE images [18]. More sophisticated methods merged images from multiple sequences for improved segmentation [19]. One major disadvantage of most of the proposed segmentation algorithms, however, is the necessary selection of a threshold value, which can be difficult because signal intensities, image homogeneity, and also image contrast depend on a wide range of sequence and image reconstruction parameters as well as hardware that may vary between vendors, scanners, sites, coils, acquisition protocols, and even subjects. This often requires manual user interaction to select an optimal threshold value for individual data sets.

In this work, we describe a fully automated algorithm for segmenting the cranial bone based on dual-echo UTE data. The algorithm is initialized by interval thresholding of an approximated T2* relaxation parameter map before applying several further, easily implementable processing steps. For one of the subjects the segmented data was also compared to images of a CT scan and used to create 3D volumetric maps of the distribution of the T2* relaxation time across the cranial bone.

Section snippets

UTE imaging sequence and protocol

A 3D UTE sequence with non-selective short hard pulse excitation and radial 3D center-out acquisition was applied to achieve ultra-short echo times below 200 μs (Fig. 1). Data sampling was switched on 20 μs before ramping up the gradients to reduce digital filtering artifacts [8]. Two mono-polar echoes were collected in a readout train. The readout re-phaser before the second echo used exactly the same gradient amplitude and timing parameters as the actual readout gradient. This design feature,

Results

In all subjects, the three cranial bone layers were clearly identifiable by their different T2* relaxation times on the T2* maps when applying a threshold interval range of 0 ms < T2* < 4 ms (Fig. 4). Inter-subject variations of the T2* distribution between red bone marrow and compact cortical bone are seen that might be age-related and additionally influenced by the different spatial resolutions. Nevertheless, all T2* distributions fall within the applied threshold range. ROI-based mean T2* values

Discussion

As the starting point for segmentation, we extracted T2* maps based on a computationally fast, linear approximation of T2* values by acquiring an ultrashort and first in-phase echo at 3 T. The obtained numerical T2* values of tissues with very short T2* relaxation times are thus most likely slightly overestimated, since finer sampling of the signal decay and non-linear fitting would be necessary to obtain more accurate T2* values. Finer sampling of echoes, however, by applying echo train

Conclusion

A fully automated algorithm for segmentation of the cranial bone from linearly approximated T2* relaxation time maps has been presented that works without user interaction for a wide range of acquisition parameters. Segmentation-derived T2* distributions were visualized as volumetric 3D renderings that allow a detailed visualization of the skull and its sutures.

Acknowledgments

This work was supported by the German Research Foundation (DFG, RE 1123/22-1) and the Interdisciplinary Center for Clinical Research (IZKF, J64) in Jena, Germany. The authors have no relevant financial conflicts to disclose with regard to this study.

References (28)

  • F. Springer et al.

    Rapid assessment of longitudinal relaxation time in materials and tissues with extremely fast signal decay using UTE sequences and the variable flip angle method

    Invest Radiol

    (2011)
  • M.D. Robson et al.

    Magnetic resonance: an introduction to ultrashort TE (UTE) imaging

    J Comput Assist Tomo

    (2003)
  • J. Rahmer et al.

    Three-dimensional radial ultrashort echo-time imaging with T2 adapted sampling

    Magn Reson Med

    (2006)
  • K.H. Herrmann et al.

    Time efficient 3D radial UTE sampling with fully automatic delay compensation on a clinical 3T MR scanner

    PLOS ONE

    (2016)
  • Cited by (0)

    View full text