The purpose of this paper is to propose a method to normalize the 1) location, 2) size, and 3) intensity of altered tracer uptake regions to allow quantitative comparison of orthopaedic SPECT/CT scans. Our method uses standardized orthopaedic frames of reference for orientation, and 3D volumetric data interpretation and thresholding to distinguish clinically relevant hot spots from background activity. Along with its introduction we strive to facilitate data collection and comparison of clinical studies using SPECT/CT. In this section we present the normalization method details. We then provide an exemplar patient comparison as a demonstration of the method's use in the Results.
The proposed normalization techniques and analysis can be carried out with any software that can reorient 3D data and quantify intensities of the SPECT voxels. We have implemented the software tools necessary for the exemplar patient comparison using a collection of open-source tools. The software was written in Python (v2.6, http://python.org/
), using the Grassroots DICOM library (GDCM v2.0.12, http://gdcm.sourceforge.net/
) to read the SPECT and CT DICOM files exported by the SPECT/CT machine, the Visualisation Toolkit (VTK v5.4.2, Kitware, http://vtk.org/
) library for surface and volume rendering, and the NumPy (http://numpy.scipy.org/
) Python library for data analysis. Note that a slight modification was required to the GDCM library to overcome an idiosyncrasy in the SPECT DICOM data for the specific SPECT/CT machine used (Symbia T16, Siemens).
Normalisation of position: orthopaedic reference frames
To reliably locate and describe the position of a region of altered metabolic activity it is essential that the reference system is well defined. Further, to overcome the relative position and orientation of the patient to the coordinate system of the scanner, the reference system should be based on anatomic landmarks that are identifiable in subsequent scans. The unique combination of anatomic and metabolic information available with SPECT/CT imaging allows the low resolution metabolic information from SPECT to be accurately localised using the high resolution anatomic information from CT.
We use well-recognised standard anatomic landmarks and corresponding frames of reference that relate to the biomechanics of the joint. The landmarks can be manually identified on a combination of CT slices and on the surface of a 3D bone reconstruction from the CT data. These landmarks define a coordinate system specific to the lower limb, independent of scanner-specific coordinates. We can then use this normalised coordinate system to localize the biomechanical and diagnostic SPECT regions of interest. Further, with this normalised coordinate system we can provide normalised views for the 3D bone reconstruction and 2D slices (i.e., "true" antero-posterior views and slices that will be the same perspective relative to the anatomy independent of the patient position and orientation in the scanner).
The femoral frame of reference is defined as containing the mechanical axis of the femur and the axis of rotation of the knee, with the origin at the centre of the joint (Figure ). This frame of reference gives the anteroposterior axis and the mediolateral axis of the femur [17
]. The midpoint of a straight line joining the surface locations of the epicondyles is taken as the centre of the knee. We use the femoral head centre as the proximal landmark defining the mechanical orientation of the femur. This point is defined as the intersection of the diameters of the head in all three planes. The mechanical axis of the femur is then defined as the line passing from the centre of the femoral head to the centre of the knee. We establish an orthogonal reference frame using the cross product of the transepicondylar line and the mechanical axis to define the anteroposterior axis, and the cross product of the anteroposterior axis and the mechanical axis to define the mediolateral axis.
Figure 1 Femoral anatomic reference frame. The femur is aligned to an anatomic reference frame, independent of patient position in the scanner, using the mechanical axis determined by the femoral head centre and the centre of the knee defined by the medial and (more ...)
The tibial orthogonal reference frame is constructed to treat the knee as a simple hinge, so that tibial sagittal planes are the same as the femoral sagittal planes (Figure ). Specifically, the mediolateral axis is taken to be the same as the femoral medioateral axis, the anteroposterior axis is defined as the cross product of the mediolateral axis and the line between the knee centre (as defined above) proximally and the centre of the talus distally, and the mechanical axis as the cross product between mediolateral axis and the anteroposterior axis.
Tibial anatomic reference frame. The tibia is aligned similar to the femur, treating the knee joint as a simple hinge. The centre of the talus is used as a distal landmark.
Normalisation of size and intensity: clinically neutral reference region
Here we propose a method for normalising the SPECT intensities between patients to allow for quantitative analysis, potentially increasing the value of SPECT imaging. Direct comparison of raw SPECT intensities is known to be problematic, due to significant variation among patients in overall uptake levels. Normalisation techniques that rely purely on the distribution of SPECT intensities, such as scaling by the maximum value, or normalising using the mean and standard deviation, are also problematic. These techniques can be biased by the fraction of tissue related voxels as compared to free air voxels (affected by patient size and field of view), the presence of an intense hotspot, or an overall inflammatory uptake response.
What we propose is to use the anatomic information provided by the CT component to establish, prior to analysis, a clinically relevant neutral region. For example, when analysing intensities relating to knee joint pain, we can define a volume in the middle of the femoral shaft (which is unrelated to the articular surface) as a reference region. The distribution of SPECT intensities from this region can be then be used to normalize other values. This approach avoids the biases associated with patient size, field of view, and normalising solely on the global distribution of SPECT intensities. It also takes advantage of the unique combination of anatomic and metabolic information offered by SPECT/CT imaging.
The optimal method to normalize hotspot intensities using reference region intensities, resulting in an analysis with the best correlation with clinical outcomes, remains an open research question and likely depends on the specific analysis conducted. Potential approaches include scaling intensities to the mean of the reference region, offsetting intensities with the mean of the reference region, using the standard deviation of the values in the reference region as a scaling parameter, or some combination of these techniques. Hotspots can then be compared in a statistical manner, using a t-test to compare the distribution of values in a hotspot with the values in the reference region. Intensity thresholds can also be defined in a straightforward way, such as more than n standard deviations away from the reference region mean.
Thresholding for volumetric 3D tracer uptake analysis and visualisation
Having established clinically relevant intensity thresholds using the method described above, we can use thresholding (an algorithmically simple operation) to distinguish altered tracer uptake voxels from background voxels. This avoids the difficult segmentation procedure normally associated with quantitative analysis of medical images. The threshold also enables simple volumetric analysis within a region. For example, one can use a rectangular volume (which is straightforward to define) to define an initial region, then use the threshold to distinguish altered tracer uptake voxels from background voxels within the region (Figure ).
Thresholding for analysing SPECT data. Using a threshold cutoff combined with an easy-to-use rectangular region selector allows for straightforward identification of hotspot voxels for further analysis.
This clinically relevant intensity also defines a boundary for the 3D surface reconstruction (i.e., isosurface rendering) of the SPECT region of interest (Figure ). The SPECT region boundary surface can be rendered with the 3D bone surface reconstruction, enabling views from any perspective including the normalized views described above. This can aid the clinician's understanding of the position of the uptake region, which is of particular use when informing a surgical procedure.
Rendering SPECT region of interest. Once a clinically relevant SPECT intensity threshold has been established, we can render a hotspot surface at that threshold for a visual appreciation of the hotspot size and position.