Template Matching in Three Dimensional Radionuclide Imaging of the by gigi12

VIEWS: 25 PAGES: 4

									       Template Matching in Radionuclide Imaging of the Lung
                John S Fleminga, Joy H Conwayb, Livia Bolta , Matthew Quintb, Amal A Salama

Departments of aMedical Physics and Bioengineering and bUniversity Medicine, Southampton General Hospital,
                                                SO16 6YD

       Abstract. Transformation of medical imaging data to a standard template of the organ of interest is a powerful
       diagnostic technique. This paper describes a method for transforming radionuclide lung imaging data to a template
       using an aligned anatomical lung outline. The template was divided into lobes and segments. The technique was
       validated using simulated images defined in the template shape. These were registered to a different lung shape and
       then back again to the template. The final doubly registered images were compared to the original. The technique
       was used in two applications. First the distribution of inhaled aerosol deposition from Single Photon Emission
       Computed Tomography (SPECT) images has been analysed by segment. Secondly, planar perfusion images
       containing segmental defects have been simulated. These were used to evaluate the errors in assessing relative left
       to right lung perfusion from the images. The registration technique was shown to produce uptake values in each
       segment with a mean error of 2.2 %. The use of the technique to analyse segmental distribution of aerosol
       deposition and produce planar images of lung perfusion with segmental defects was demonstrated. The technique is
       considered a useful tool in the analysis and interpretation of radionuclide lung imaging

1 Introduction
Elastic transformation of medical imaging data to a standard template of the organ of interest is proving a very
powerful analytical tool, enhancing objectivity of image interpretation [1]. Such techniques have had limited
application to lung imaging. They have, however, been used to analyse Single Photon Emission Computed
Tomography (SPECT) images of the distribution of inhaled aerosol in the lung [2]. The lung data are first
transformed to a spherical shape using aligned anatomical images (computed tomography (CT) or magnetic
resonance imaging (MRI)) to define the lung outline. A spatial model of the airway tree can then be used to
allow the deposition of aerosol in different airway generations of the lung to be calculated [3]. Although useful,
the spherical transformation has the limitation of not reflecting the real asymmetric segmental structure of the
lung. The ability to transform to a real lung shape would give the opportunity of describing the variation of
radioaerosol deposition by segment. This would have two particular areas of application. (i) It would provide
experimental information on the distribution of inhaled aerosol in the different lung segments, which would be
valuable to computer modellers of deposition in evaluating their theoretical predictions. (ii) Segmental lung
models have been used in simulating radionuclide images of lung ventilation and perfusion. This has helped in
understanding and interpreting these images. However current applications have only used a single shape model
[4]. Elastic transformation would allow the segmental model to be transformed to a variety of different shapes
and hence enable modelling of more realistic anatomical variability.

This paper first describes the definition of a segmental lung model. A technique for elastic transformation of a
three-dimensional radionuclide image of the lung to this standard template is then presented. The transform
technique is validated using simulated gamma camera images. Finally applications of the technique in
analysing the distribution of aerosol in the different segments of the lung and in producing simulated data for a
national audit of radionuclide lung imaging are described.

2 Methods

2.1 Definition of Segmental Lung Template

A segmental model of the lung has been developed based on the lung shape of a male subject with normal lung
function and a near average value of lung size as adjudged from the functional residual capacity (FRC). MR
images of the subject’s thorax were segmented to obtain the lung outline and then each voxel in the lung
assigned a value according to the lung segment to which it belonged; there are ten segments in each lung. This
was achieved by digitising photographs of a cast of the lung airway in which the different segments were
denoted by different colours, and registering them to the outline of the standard lung model. Figure 1 shows an
example slice through the model.
     Figure 1. Example slice through the segmental model showing different segments by their grey levels

2.2 Transform Technique

The elastic transformation of the SPECT imaging data to a standard template was achieved using a spherical
transform based on definition of the lung outlines of the subject under investigation and of the standard lung
template. This was preferred to more complicated schemes, such as active shape models, for its simplicity. The
transform also required definition of a single common anatomical location in both the subject and template. The
hilum of the lung, where the main bronchus enters the lung space, is in many respects the natural choice for the
centre of a spherical transform, as both airways and blood vessels branch out approximately radially from this
point. However the hilum is either on, or close to, the edge of the lung and therefore cannot be used, as nearly
half the radial paths from it have a very short or zero length within the lung. Therefore a point in the centre of
the lung O(ox, oy, oz) is chosen, which is related to the hilum position H(hx, hy, hz) such that:

          ox = (hx + px ) /2,     oy = hy,        oz = hz                         (1)

where x, y and z represent the left/right, anterior/posterior and superior/inferior directions respectively. p x is the
x position of the lateral edge point on the lung at the same y and z co-ordinates as the hilum. The transform is
illustrated in Figure 2. The purpose is to determine, for each voxel in the template shape, the corresponding
position in the subject shape and then transfer the value of the SPECT image in the subject voxel to the
equivalent template voxel. Each voxel Vt in the template shape is characterised by its direction cosines and the
fractional radial distance (ft = OVt / OEt) from the central point (Ot) to the equivalent extrapolated position on
the edge (Et). The line OEs in the subject shape is defined as that passing through O s with the same direction
cosines as OVt . For each voxel along this line the fractional distance to the edge (fs = OVs / OEs) is calculated.
The voxel, Vs, with the value of fs nearest to ft, is found. The SPECT image value in this voxel is transferred to
voxel, Vt, in the template shape. This process is repeated for all the template voxels.




                                                                   Ot
                                  Os                        Ht               Pt
                      Hs                     Ps
                                                                     Vt
                                  Vs
                                                                             Et
                                           Es




                           Subject shape                    Template shape


                     Figure 2. Schematic diagram illustrating the principle of the registration

2.3 Validation of Registration using Simulation

The technique was validated using simulated data. A three-dimensional image, containing a uniform
distribution of activity in all segments except one, in which there was no activity, was defined for the template
shape. This was registered to a different lung shape obtained from MR imaging of another subject, giving an
estimated activity distribution with respect to the new shape. The reverse process was then applied to register
the image back to the template shape and the average count in each of the template segments obtained. In
particular the count in the segment with zero activity was noted. Since the transformation is not exact, voxels on
the edge of segments will be misplaced in the process. The fact that it is being carried out twice will mean that
registration errors are likely to be double that of a single transformation. This experiment demonstrated how
segmental volumes of interest were affected by the transform. In order to study the effect on SPECT data, a
further evaluation study was performed on simulated SPECT images of the above activity distribution. These
images were subjected to the two way registration process described above and the activity in the segment with
zero true activity was noted before and after the registration. The process was repeated using two different
segments and two different lung shapes.

2.4 Clinical Application

Two areas of clinical application have been initiated. The first is the evaluation of the segmental distribution of
inhaled aerosol. Three-dimensional radionuclide imaging using SPECT and PET (Positron Emission
Tomography) has been used to assess the distribution of inhaled aerosol distribution in the lung. The
registration technique has been applied to example SPECT images with a view to assessing the regional
distribution of aerosol by segment. This is achieved by calculating the mean activity per unit volume deposited
in each segmental volume of interest in the registered SPECT images.

The second area of clinical application is in the simulation of images for a national audit of analysis of
radionuclide lung ventilation/perfusion images. These images are often quantified to produce estimates of the
relative function of each lung. The audit process is aiming to investigate the reliability of software used to
process such studies. The use of simulated data enables the results of analysis to be compared to the true known
values of relative function used in the simulation and therefore estimates of both accuracy and precision may be
obtained. Simulation of segmental defects was obtained by creating a true distribution of activity in the
template in which one or more of the segments is set to zero while all the others contain a normal level of
uptake. This was then registered to a different lung shape obtained from MR imaging of a different subject.
Anterior and posterior planar images of the registered distribution were then simulated assuming typical
imaging conditions used in clinical practice. A set of ten such image pairs were produced and then analysed for
relative right/left lung function. The results were compared to the true values assumed for the simulation.

3 Results
Figure 3 shows an example of a theoretical image of perfect resolution before and after the two-way registration
experiment. It is clear that the registration was not exact as a few voxels near the edge of the segment
containing zero counts are malpositioned following transformation. The figure also shows the results of
transformation of the same slice of the equivalent simulated gamma camera data, which incorporates realistic
noise and resolution. The images are very similar in appearance. The mean count in the segment with
theoretically no activity was 26% of that in the surrounding lung due to the influence of the partial volume
effect. This value after the two way registration process is very similar at 27% indicating that although the
registration was not perfect it does adequately deal with the task of registering these relatively poor resolution
SPECT images. The mean difference in count in this segment, between images assessed before and after
registration, for the segment with no activity was 2.2%.




    Figure 3 The two images on the left are example slices of a theoretical distribution of activity in the lung in
    which one segment is set to zero before and after the two-way registration experiment. The two images on
                  the right are of the same slices of the corresponding simulated SPECT images.

Figure 4a shows an example of a transverse slice from a SPECT image of a subject who has inhaled a
radiolabelled aerosol from a nebuliser. The slice is towards the base of the lung and illustrates the greater
concentration found at the posterior base. The segmental analysis allows the deposition to be quantified. The
deposition in the posterior basal and apical superior segments were about twice that in the surrounding
segments. Figure 4b shows examples of simulated posterior planar images of lung perfusion, which are to be
used for audit of ventilation/perfusion quantification software. The images demonstrate the variety of lung
shapes, sizes and distribution patterns that can be simulated. All ten assessments of relative perfusion from the
simulated image analysis were within one percentage point of the true value.
               a                        b


Figure 4 (a) Example transverse slice of a SPECT image of a distribution of an inhaled aerosol in the lung of a
  normal subject. (b) Example simulated planar gamma camera images of perfusion for use in evaluation of
                                            quantification software

4 Discussion and Conclusions
Elastic registration to a standard template is proving a very powerful technique in medical image analysis. This
paper has described a method for registration of SPECT images of the lung. The method has been validated in a
two-way registration experiment. This showed that the registration was not perfect on a voxel by voxel basis,
but that when applied to simulated SPECT data that modelled real SPECT image quality, quantification of mean
activity in a segment was achieved with a good level of precision. The method has the advantage over the
spherical transform of providing the data in a recognisable lung shape. It has allowed image data of aerosol
deposition within the lung, to be quantified by segment. This will be valuable to modellers of deposition
mechanisms in the lung, who require detailed experimental data to help validate and develop their models.
Finally the method is shown to be of value in improving simulation of lung images for evaluation of image
interpretation and quantification. The example given in this paper shows how the technique can be used to
evaluate quantitative techniques. This data is currently being circulated to different hospitals around the country
by the Nuclear Medicine Software Working Party of the IPEM to enable a national audit of techniques for
quantification of ventilation/perfusion imaging. Most previous audits have involved the distribution of patient
images. The interpretation of these studies has been limited by the fact that the true answers were not known,
allowing only precision between centres to be determined. Basing the new lung audit on simulated data means
that accuracy as well as precision can be evaluated. Magnussen et al [4] have used simulation of the gamma
camera imaging process for evaluating interpretation of ventilation/perfusion imaging based on a segmental
model of the lung. However their method only allows images of one particular lung shape to be produced. The
technique described in this paper shows how this concept can be extended to allow simulation of different image
shapes. It should be noted however that there is still a limitation of the method in that the segmental model is
generic. It is known that there is considerable individual variability in segmental anatomy and the current
technique does not take this into consideration. It is possible that such detailed individual anatomy may be
available in the future with improved CT imaging or the use of hyperpolarised helium MRI.

In conclusion a technique for template registration of the lung has been described, which has promising
applications in analysis and interpretation of radionuclide lung imaging. Many of the ideas described in this
paper could readily be applied to other types of radionuclide imaging and other imaging modalities.

Acknowledgements

The authors would like to thank AstraZeneca and the Chartered Society of Physiotherapy Research Foundation
for their financial support of this work.

References

1. K J. Friston, A.P. Holmes, K.J. Worsley, J.P. Poline, C.D. Frith, R.S.J. Frackowiak. “Statistical parametric maps in
    functional imaging”. Human Brain Mapping 2, pp 189-210, 1995
2. J.S. Fleming, P.M. Halson, J.H. Conway, E.A. Moore, M.A. Nassim, A.H. Hashish, A.G. Bailey, S.T. Holgate, T.B.
    Martonen. “Three dimensional description of pulmonary deposition of inhaled aerosol using data from multimodality
    imaging”. Journal of Nuclear Medicine, 37, pp 873-877, 1996
3. J S. Fleming, J.H. Conway, S.T. Holgate, A.G. Bailey, T.B. Martonen. “Comparison of methods for deriving aerosol
    deposition by airway generation from three-dimensional radionuclide imaging” Journal of Aerosol Science. 31, pp 1251-
    1259, 2000.
 4. J. S. Magnussen, P. Chicco, A.W. Palmer, D.W. Mackey, M. Magee, I.P.C. Murray, G. Bautovich, K. Allman, G. Storey,
    H Van der Wall. “Enhanced accuracy and reproducibility in reporting of lung scintigrams by a segmental reference
    chart”. Journal of Nuclear Medicine 39, pp 1095-1099, 1998.
5. A.S. Houston, D.R. Whalley, J.V. Skrypnuik, P.H. Jarritt, J.S. Fleming, P.S. Cosgriff. “UK audit and analysis of
    quantitative parameters obtained from gamma camera renography.” Nuclear Medicine Communications. 22, pp 559-566,
    2001.

								
To top