Elsevier

NeuroImage

Volume 25, Issue 1, March 2005, Pages 278-286
NeuroImage

Atlas-guided non-uniform attenuation correction in cerebral 3D PET imaging

https://doi.org/10.1016/j.neuroimage.2004.11.021Get rights and content

Abstract

Photon attenuation in tissues is the primary physical degrading factor limiting both visual qualitative interpretation and quantitative analysis capabilities of reconstructed Positron Emission Tomography (PET) images. This study investigates the implementation and applicability of transmission atlas-guided attenuation correction in cerebral 3D PET imaging, thus eliminating the need for acquisition of a measured transmission scan. Patient-specific attenuation map is derived by anatomic standardization through nonlinear warping of a stereotactic transmission template obtained by averaging 11 scans of normal subjects. This template is coregistered to a specially designed tracer-specific 18F-[FDG] template constructed by scanning 17 normal subjects in resting condition during tracer uptake in a dark room. This emission template is first coregistered and spatially normalized to preliminary PET images of subjects corrected for scatter and attenuation using an approximate calculated method. The resulting transformation matrices are recorded and re-applied to the transmission template. The derived attenuation map is then forward projected to generate attenuation correction factors to be used for correcting the subjects' PET data. Twelve cerebral clinical studies are used for evaluation of the developed attenuation correction technique as compared to the standard pre-injection measured transmission-based method used in clinical routine. Statistical Parametric Mapping (SPM2) analysis is used to assess significant differences between images obtained using both techniques. The subjective qualitative assessment shows no significant visual differences between atlas-guided and transmission-based attenuation correction methods. However, the quantitative voxel-based analysis comparing atlas-guided to transmission-based attenuation corrections suggest that regional brain metabolic activity increases significantly bilaterally in the superior frontal and precentral gyri, in addition to the left middle temporal gyrus and the left frontal lobe. Conversely, activity decreases in the corpus callosum in the left parasagittal region. A new non-uniform attenuation correction method is thus proposed, which is suitable for both research and clinical routine applications in 3D brain PET imaging on a transmissionless PET scanner or when a patient-specific transmission scan is not available.

Introduction

Functional brain imaging using single-photon emission computed tomography (SPECT) and positron emission tomography (PET) has steadily gained importance both in clinical setting and research environment over the past few years. Reliable subjective qualitative visual interpretation and accurate objective quantitative analysis of functional brain images rely heavily on the intrinsic performance parameters of the PET scanner and the computational models used to correct the data for physical degrading factors. There is in general a consensus among the nuclear medicine community that photon attenuation is the dominant physical degrading factor besides contribution from scattered photons and other associated errors (Zaidi and Sossi, 2004). The scientific community has witnessed an impressive development of computational models for accurate attenuation correction in PET imaging as an increasing number of clinical indications and research applications are being explored.

The accuracy achieved by the attenuation correction procedure is evidently dependent on the rigour followed to derive the attenuation map. Two broad categories of techniques have emerged: calculated methods, which are based on an assumed anatomical model representing the shape and spatial distribution of attenuation coefficients in the head and measured methods, which in general rely on an additional acquisition of a transmission scan using either positron-emitting (68Ga/68Ge) or single-photon emitting (137Cs) radionuclide sources or an X-ray tube on recently developed dual-modality PET/CT imaging systems (Zaidi and Hasegawa, 2003). It has been shown that motion-induced misalignment between pre-injection transmission and emission scans can result in erroneous estimation of regional tissue activity concentrations (van den Heuvel et al., 2003). More recently, a new method based on coregistered segmented magnetic resonance imaging (MRI) was proposed for brain imaging (Zaidi et al., 2003). The main motivation behind the development of this method is to achieve accurate attenuation correction with a transmissionless prototype brain PET tomograph. Removal of the transmission scanning component simplifies greatly the complexity of the tomograph's design and acquisition protocols and contributes significantly to lowering the cost of the prototype and the radiation dose delivered to patients and healthy subjects. In fact, the authors are involved within the Computed Imaging for Medical Applications (CIMA) collaboration in the design of a novel and innovative high-resolution Compton-enhanced 3D brain PET tomograph dedicated to brain research1 (Braem et al., 2004). This concept leads to an image reconstruction which is free of any parallax error and provides a uniform spatial and energy resolution over the whole sensitive volume. Ideally, it was planned that a 3D brain T1-weighted MRI should be acquired for all subjects participating to research protocols prior to the PET scan. It was realized afterwards that for logistic and practical organizational reasons, the MRI is not always available on the hospital picture archiving and communication system (PACS) and thus another alternative for accurate attenuation correction should be sought.

Historically, empiric calculated methods for deriving the attenuation map in functional brain imaging such as the uniform fit-ellipse method and the automated detection of the head contour followed by assignment of known attenuation coefficients (brain and surrounding skull) were proposed and implemented in software supplied by scanner manufacturers to end-users (Zaidi et al., 2004). Their limitations led to the development of more sophisticated models including the automated method computing a 3-component attenuation map (Weinzapfel and Hutchins, 2001) and the inferring anatomy from a head atlas approach (Stodilka et al., 2000). The former generates an estimated skull image by filtered backprojection of the reciprocal of an emission sinogram whose thickness and radius are estimated from profiles extracted from this image. Whereas the latter derives the attenuation map by registering the brain component of a digital head phantom of a single subject (Zubal et al., 1994) with a source distribution having appropriately scaled specific activities of the gray matter, white matter, and ventricles, to a preliminary PET reconstruction and then applying the resulting spatial transformation to the full head atlas. As reported by many investigators, the major limitation of the Zubal head phantom is that the sinus is larger than usual (Arlig et al., 2000, Stodilka et al., 2000, Zaidi et al., 2004). The method presented in this work further extends this approach and improves its robustness by constructing both transmission and tracer-specific emission atlases based on average patient populations rather than relying on an atlas based on a single subject and a hypothetical tracer distribution. The second improvement is related to application of nonlinear warping for anatomic standardization of stereotactic templates and patient images in contrast to simple global rescaling (7 parameters model) procedures used in the previous studies. In the narrow sense of the word, there is a major conceptual difference between anatomic standardization—also called spatial normalization—and coregistration (Senda, 2000). Basically, coregistration aims to match images of a single subject, usually of a different tracer or modality, through a rigid-body transformation. On the other hand, the purpose of anatomic standardization is to transform brain images of individual subjects into a standard brain. Another important difference is that the true solution exists for registration but not for anatomic standardization (Van Laere and Zaidi, 2005). Standardizing patient brain images should therefore be performed with caution (Senda, 2000). The performance of the proposed attenuation correction algorithm is evaluated using clinical data through comparison to the standard procedure used in clinical routine based on measured pre-injection transmission scanning.

Section snippets

Description of algorithm

The proposed attenuation correction algorithm uses an emission and transmission atlases and preliminary PET reconstructions of the subject's brain images to construct a non-uniform attenuation map adapted to the subject anatomy. The main advantage being that the method does not require additional acquisition of a transmission scan and does not rely on the questionable assumption of perfect coregistration between the transmission and emission scans when using a pre-injection transmission

Results

A typical plane of a patient brain attenuation map obtained with measured pre-injection transmission and atlas-guided methods are shown in Fig. 3. Horizontal profiles through the sinus cavities and middle of the slice are also shown to demonstrate quantitatively the differences between the different techniques. Note that the typical profile generated by the calculated fit-ellipse method is shown to highlight the limitations of this approach often used in clinical routine in the sense that

Discussion

There are many physiological, biochemical and pharmacological functions of the human brain that can be explored with PET imaging, using for example tracers as 18F-[FDG] to measure the metabolism of glucose, 15O-[H2O] to assess the rate of blood flow or a large and increasing library of 11C-labeled radioligands to characterize neuroreceptor function. It should be emphasized, however, that many of these applications rely on a solid quantitative foundation, which is highly dependent on the

Conclusion

A new attenuation correction method for 3D brain PET guided by transmission and tracer-specific emission atlases constructed from normal databases representative of average patient populations has been proposed. The method is suitable for either clinical routine and research applications in 3D brain PET imaging on a transmissionless PET scanner or when a transmission scan is not available. Moreover, the removal of the usual pre-injection transmission scan contribute to overall reduction of

Acknowledgments

This work was supported by the Swiss National Science Foundation under grant SNSF 3152A0-102143, the National Center of Competence in Research CO-ME and Geneva University Hospital R and D funds under grant PRD-04-1-08. This paper is dedicated to the memory of Prof. F. Terrier (Head of the Department of Radiology, Geneva University Hospital) who sadly passed away this summer.

References (29)

  • G.W. Goerres et al.

    Head and neck imaging with PET and PET/CT: artefacts from dental metallic implants

    Eur. J. Nucl. Med. Mol. Imaging

    (2002)
  • K. Ishii et al.

    Statistical brain mapping of 18F-FDG PET in Alzheimer's disease: validation of anatomic standardization for atrophied brains

    J. Nucl. Med.

    (2001)
  • P.E. Kinahan et al.

    Analytic 3D image reconstruction using all detected events

    IEEE Trans. Nucl. Sci.

    (1989)
  • S. Minoshima et al.

    Anatomic standardization: linear scaling and nonlinear warping of functional brain images

    J. Nucl. Med.

    (1994)
  • Cited by (70)

    • PET/MRI attenuation correction

      2022, Biomedical Image Synthesis and Simulation: Methods and Applications
    • MR Imaging-Guided Attenuation Correction of PET Data in PET/MR Imaging

      2016, PET Clinics
      Citation Excerpt :

      A summary of the pros and cons of the atlas-based MR-AC approaches is shown in Box 2. The first atlas-based method used the well-known Statistical Parametric Mapping, University College London (SPM2) to spatially normalize the PET image of the subject to the PET image of an atlas and the corresponding atlas transmission image.27 The transmission image was then warped back into the (PET) subject space using the inverse deformations estimated during the spatial normalization process.

    • Future image acquisition trends for PET/MRI

      2015, Seminars in Nuclear Medicine
    • Simultaneous PET and MR Imaging of the Human Brain

      2023, Functional Neuroradiology: Principles and Clinical Applications
    View all citing articles on Scopus
    View full text