Evaluation of ZTE localization accuracy for PET/MR hardware attenuation correction
Gaspar Delso1, Mohammad Mehdi Khalighi2, Sabrina Epp3, Felipe de Galiza Barbosa3, Tetsuro Sekine3, Edwin ter Voert3, and Patrick Veit-Haibach3

1GE Healthcare, Zurich, Switzerland, 2GE Healthcare, Stanford, CA, United States, 3University Hospital, Zurich, Switzerland

Synopsis

One challenge of PET/MR imaging is the correction of photon attenuation caused by hardware in the field-of-view. For the particular case of MR local coil attenuation, a solution based on stored templates is provided by commercially available clinical PET/MR systems. This solution, however, is limited to rigid coils docked in a pre-defined position. A more general method encompassing other coil types would be desirable to improve the accuracy of PET images. The goal of the present study is to investigate the accuracy of fast ZTE acquisitions for the correction of local coil attenuation.

Purpose

One of the main challenges of PET/MR imaging is the correction of photon attenuation, caused by the patient as well as any hardware in the field-of-view. For the particular case of MR local coil attenuation, a solution based on stored templates is provided by both commercially available clinical PET/MR systems (Siemens mMR and GE SIGNA). This solution, however, is limited to rigid coils docked in a pre-defined position (typically the head/neck coil). A more general method encompassing other coil types would be desirable to improve the accuracy of PET images1-4. A recent development in MR-based attenuation correction is the use of 3D zero echo time (ZTE) imaging for bone tissue identification5. This sequence provides high-resolution images of fast-decaying species without requiring preparation pulses or multiple echoes, making it a very time-efficient acquisition. An interesting property of ZTE is that it is also capable of capturing the signal from certain components within the local coils, which could be used as landmarks to register an attenuation template. The goal of the present study is to investigate the accuracy of fast ZTE acquisitions for the correction of local coil attenuation.

Methods

A fast ZTE acquisition, compatible with clinical workflow, was set up at the SIGNA PET/MR system of the University Hospital of Zurich. The sequence was tested for the localization of the 8-Channel Breast Coil, for which an attenuation template is included in the system. The acquisition parameters were: FOV 28cm, ST 2.5mm, 100 slices, FA 1°, frequency 112, NEX 2, BW 62.5kHz, acquisition time 20s. A post-processing algorithm was implemented in Matlab (The MathWorks, Inc., Natick, MA) to eliminate all patient tissue from the acquired images and segment the visible coil components. A set of features were automatically extracted from each segmented component, including volume, centroid, principal axes and dimensions. The point clouds defined by the centroids of all identified components were rigidly registered to evaluate the consistency of each component’s localization and determine the optimal set of landmarks for template alignment. The repeatability of coil localization was tested first on 10 phantom acquisitions without repositioning, and then on 10 acquisitions where the whole phantom setup, landmarking and acquisition procedure was performed. The impact of realistic coil loading conditions was evaluated on seven acquisitions of human subjects (3 healthy volunteers and 4 patients).

Results

When only the repeatability of the pulse sequence and post-processing were considered, the segmented landmark centroids were in all cases obtained with sub-millimetric precision: range [(-0.2, -0.3, -0.7)mm, (0.4, 0.1, 0.4)mm], standard deviation < (0.2, 0.2, 0.5)mm. When the entire setup and landmarking procedure was included, the precision dropped significantly: range [(-15.3, -2.4, -7.1)mm, (7.4, 2.8, 8.9)mm], standard deviation ϵ [(0.3, 0.4, 0.2)mm, (5.7, 3.8, 3.5)mm]. Similar results were obtained on human subjects: range [(-10.6, -25.8, -6.5)mm, (14.9, 9.3, 11.1)mm], standard deviation ϵ [(0.3, 0.5, 0.1)mm, (8.5, 13.2, 6.6)mm]. On phantom acquisitions, the position of the coil could be registered to a common reference with a standard deviation of (0.5, 1.2, 1.0)mm and (0.9, 0.2, 0.9)°. On human subjects, the deviation was (1.5, 1.3, 1.3)mm and (0.6, 0.6, 0.5)°.

Discussion

In general, the majority of landmarks were located with a precision oscillating between 1 and 3mm, with occasional outliers caused by mis-segmentation. After selection of the optimal set of landmarks, the precision of coil positioning is well below the typical voxel size of PET attenuation maps (~4mm), indicating that ZTE imaging would be suitable for a template-based attenuation correction approach. Alternative approaches based on MR-visible landmarks (e.g. oil pills or micro-coils) placed on the local coil have been published in the past, but they require hardware changes and could be an issue in case of phase wrap-around. A limitation of the present study is the reduced number of clinical cases evaluated. New patients are currently being recruited through an ongoing clinical study. Future work will be aimed at extending the method to other local coils (e.g. GEM anterior array and Flex suite)

Conclusion

The results suggest that proton density-weighted ZTE imaging can be effectively used for local coil positioning in template-based attenuation correction.

Acknowledgements

No acknowledgement found.

References

1. MacDonald, et al.,Med Phys 2011; 38: 2948-2956. 2. Paulus, et al. Med Phys 2012; 39: 4306-4315. 3. Wollenweber et al., Magn Reson Mater Phy. 2013; 27:149-159. 4. Kartmann, et al., Med Phys 2013; 40. 5. Wiesinger et al., ISMRM 2014; 5819.

Figures

Fig. 1 - Volume rendered lateral views of two typical ZTE datasets, showing the different elements of the coil’s internal circuitry that can potentially be used as landmarks for the registration of a template.

Fig. 2 - Volume rendered frontal overlay of a CT scan of the 8-channel breast coil (in white) and a typical ZTE MR acquisition (in red), showing which coil components can be identified in the images.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3560