European Nuclear Medicine Guide
Nuclear Medicine is, in general, a quantitative technique allowing the determination of the in-vivo absolute concentration and associated distribution of activity in a given region of interest (ROI). However, prior to the extraction of such quantitative information, there are a number of errors that need to be accounted for and are predominantly associated with the physical principles governing the overall detection process (attenuation, scatter), the performance of the imaging devices (resolution, dead time), and parameters related to acquisition protocols and patients (e.g. respiratory, involuntary motion, and tracer kinetics).
Due to the interaction of photons with tissue, there are different effects that need to be accounted for in order to accurately recover the in-vivo activity concentration at different locations within the body. The two major types of interaction, namely photoelectric absorption and Compton scatter effects are responsible for photon attenuation leading to Ax=Ao e-μd, where Ax is the recovered activity, Ao is the emitted activity at a depth d from the detector surface, and μ is the tissue linear attenuation coefficient proportional to the tissue density. Photon attenuation depends on the patient size and the region to be imaged. In addition, since the composition of the tissue varies as a function of the body location, the errors introduced as a result of attenuation varies from one region to the next. For example, the attenuation errors associated with the quantitative activity concentration recovery are more important for “deeper” located lesions than those located more superficially. Photon attenuation is also more significant in single photon imaging relative to positron emission tomography (PET) given the lower associated photon energies. As an example, in the case of 99mTc-brain imaging, the reduction in the number of detected counts can be over a factor of 4 because of attenuation effects. Similarly, it is more challenging to perform attenuation correction in single photon imaging compared to PET, where the attenuation effects are not dependent on the depth of the emission within the body but only on the body size. Historically, due to the lack of appropriate hardware and software solutions in addition to exam time issues, attenuation correction has not been widely used in single photon imaging, while it has been regularly applied in PET. In today’s era of multimodality devices, the use of anatomical images (CT, MRI) are routinely used for the attenuation correction of both PET and SPECT images. Within this context, different parameters must be taken into account to reduce any potential errors associated with the use of the anatomical images for attenuation correction. These include differences in the acquisition protocols for nuclear and CT (or MR) images, presence of oral/intravenous contrast agents or metallic implants, accurate attenuation parameters’ derivation and scaling, and the presence of CT/MR image artefacts.
In terms of scatter, Compton interactions cause a deviation in the photon’s direction and not only reduce the number of detected events but also cause an increase in the noise based on the detection of some of these scattered photons and associated coincidences in PET imaging. More specifically, the number of these detected scattered events depends on the amount of tissue present as well as its density. Despite an associated reduction in the energy of these detected photons, some of them are still detected within the energy windows used both in single photon and PET imaging. On the one hand, the energy resolution of single photon imaging systems has been consistently better than that of the PET devices, which in turn facilitates the use of energy window based approaches for scatter compensation. On the other hand, the most popular scatter correction methodology in PET imaging is based on modelling the scattered coincidences’ contribution for those events having undergone a single Compton scatter interaction prior to their detection.
Other detection system related parameters that need to be accounted for in order to improve overall image quantitative accuracy include dead time effects and spatial resolution. Dead time effects are mostly encountered in high activity concentration environments, for example during the passage of the initially injected activity bolus in combination with dynamic imaging protocols (see section in dynamic imaging below). It is quite straight forward to correct for such dead time effects, if the count rate characteristics of the corresponding detector systems are known.
In the case of spatial resolution, it is well known that nuclear medical imaging is globally characterized by poor spatial resolution in comparison to anatomical imaging modalities. As a result, for objects smaller than twice the spatial resolution of the imaging device, the activity spreads out beyond the physical limits of the object. This observation is known as the partial volume effect, and it is a major obstacle in accurately recovering the activity concentration in small regions of interest. Clearly, this effect is more significant in the case of single photon relative to PET imaging given the superior spatial resolution associated with PET. In addition, any approach to automatically segment regions of interest in the reconstructed images need to account for these partial volume effects, which make regions of interest look bigger than what they physically are. A simple way is to sufficiently enlarge the drawn region of interest in order to recover all of the activity present, but the average activity concentration will be subsequently lower than the real one. A different approach is based on the use of recovery coefficients which can be used to multiply the region of interest based image derived activity concentrations with a correction factor accounting for the partial volume effects. These recovery coefficients are obtained by imaging “hot spheres” phantoms containing different size spheres of one to a few centimetres in diameter. These spheres are filled with different activity levels, while the phantom’s background is uniformly filled in order to lead to different contrast levels. Although these recovery coefficients can lead to an improved quantification, they are based on certain assumptions such as the use of spherical-like region of interest analysis and the presence of a uniform background activity concentration. In the last few years, different approaches have been developed in both SPECT and PET imaging for the integration of the system resolution during the image reconstruction process and enables an improvement in both qualitative and quantitative image accuracy. In the case of SPECT imaging, this modelling also accounts for the septal penetration and associated collimator scatter effects.
Finally, one has to consider parameters associated with the patient and associated physiological motion in order to reduce impact on the quantitative accuracy of the reconstructed images. In terms of physiological motion, two different components need to be accounted for: respiratory motion and cardiac beating. Both of these can influence thoracic and cardiac imaging. In order to account for both types of motion, different external monitoring devices need to be employed. For respiratory motion, a pressure belt, a spirometer, or monitoring the motion of reflective points attached to the patient chest as a function of respiration can be used. In the case of the cardiac motion an ECG monitor can provide the necessary cardiac beating signal. The physiological signals derived from these devices can be subsequently used to trigger/synchronize the data acquisition leading to the construction of data frames corresponding to particular respiratory phases or amplitudes and cardiac beating state (diastole, systole). It is worth noting that alternative approaches have been recently developed that would allow for the physiological motion information to be derived directly from the raw data. These physiological motion aspects of the patient may be particularly demanding in the case of multimodality imaging given the different acquisition conditions used by various imaging devices. In the case of CT imaging, acquisitions over the lung fields are usually performed with the patient blocking their respiration, while in the case of nuclear imaging, the acquisition length is only compatible with a respiratory motion average protocol. Such differences will not only cause qualitative mismatches in the superposition of the anatomical and functional images, but they will also cause quantitative errors in the use of the anatomical images for attenuation correction of the functional images. These errors and associated artefacts may be more significant in the field of PET imaging for lung cancer and PET/SPECT cardiac imaging applications.
Finally, another form of motion that should be accounted for is that of the tracer. Most nuclear medicine protocols concern static acquisitions that are realized a certain time after the tracer injection so as to facilitate its accumulation within the body and in regions/organs of interest. The assumption behind this is that the injected tracer has reached an equilibrium within the area/organ of interest by the time of the acquisition, and therefore the activity concentration remains constant throughout this static acquisition, albeit there are physical and biological decay issues that can be easily accounted for. However, there are certain cases where the study of dynamic phenomena requires an acquisition over different phases of the tracer accumulation process. The quantitative analysis of such dynamic acquisitions is described in the dynamic imaging section below.
Having corrected for all physical and patient specific variables, the reconstructed images can be analysed to allow the extraction of quantitative physiological parameters. These parameters can be fully quantitative or semi-quantitative as is the case in most static image acquisition protocols. Moving beyond a purely visual image interpretation involves, at minima, the definition of a region of interest over which the concentration of activity will be recovered using the reconstructed static images. This ROI can be drawn manually based on knowledge of the physical properties of the image formation process but also of the underlying physiological properties. An alternative involves the use of simple thresholding approaches which define a ROI based on a predetermined level of activity concentration that is assumed related to a diseased state. In this case, all image voxels with an activity concentration over this threshold will be included in the semi-automatically defined ROI. This approach is known as a fixed-threshold based image segmentation. More sophisticated automatic segmentation approaches for the determination of a ROI in a 2D image or a 3D volume of interest (VOI) exist and may be available in different commercial software platforms used today in nuclear medicine. These are based on different approaches (for example statistical analysis of image intensity histogram distributions) that allow classification of a group of voxels with a given level of intensity (activity concentration) to a given ROI. For example, in PET oncology applications such segmentation approaches can be used to automatically define the 3D tumour volume from the reconstructed PET images. Following the definition of a given ROI, the recovered activity concentration can be normalized using different pertinent patient specific parameters (for example patient weight and/or injected activity) which allows inter-patient comparisons but also the grouping of different patients having the same disease within uniform patient cohorts. Classical semi-quantitative image parameters in nuclear medicine concern the PET image derived standardized uptake value (SUV) commonly used in clinical oncology studies and defined as the activity concentration normalized by the patient weight and injected activity. In this case the activity concentration can be either defined by the voxel with the highest (SUVmax) or the average (SUVmean) activity concentration value within a ROI (defining for example a tumour for oncology applications). Both the functional 3D volume but also the SUVs have been shown to have significant predictive and prognostic value in PET imaging.
Once the 3D volume has been defined it is also possible to go further and analyse the activity distribution in order to identify any heterogeneity characteristics that may be present. The hypothesis behind such an analysis is that the image heterogeneity may be related to underlying tissue/cellular heterogeneity. In order to derive such parameters, simple metrics can be used by looking at the symmetrical properties of the intensity histogram of the voxels contained with this 3D volume. More sophisticated quantitative measures can be obtained by performing texture analysis in order to characterize any heterogeneities present in the activity distribution not only at a global level (histogram analysis) but also on local and regional levels. All of these quantitative parameters are commonly known today as radiomics.
Dynamic scanning is a very popular acquisition protocol in nuclear medical imaging for different applications. This acquisition mode can be used both in planar single photon imaging (2D + t) as well as in the SPECT/PET tomographic imaging mode (3D + t). The time information is used to bin the data in different temporal frames which can be either visualized in cine-mode or quantitatively analysed in order to obtain what is known as a time activity curve (TAC). A TAC can be derived by placing a ROI over a specific area/organ of interest and plotting the variation in the ROI image count density over time. If there is significant underlying activity concentration within the ROI, the recovered count density needs to be background corrected. Such a correction can be performed by selecting a background ROI in an area close to the organ of interest, where it is expected that the count density will represent the background activity concentration levels. Clearly, when one draws these ROIs, one needs to account for the partial volume effects described in the previous sections above. This is particularly critical when the area of interest is small and close to other rapidly varying activity concentration regions. Once corrected for background activity, the TAC represents the variation in the activity concentration within the area/organ of interest and, as such, can be used in different fashions to derive fully quantitative indices. The simplest is based on the variability of these TACs between healthy and diseased states. More quantitative physiological parameters can be derived by modelling the tracer kinetics. In general, such a modelling process considers different parts of the organ of interest as a set of compartments (physical or biochemical spaces) assuming a uniform activity concentration. Different variable rates (kinetic parameters) and associated differential equations are used to denote the tracer exchanges between the compartments used to describe the function of a given organ of interest. Ideally these kinetic parameters are determined by fitting the measured TACs with curves based on the differential equations describing the function of a kinetic model. Finally, these models can be also used to derive what are known as parametric images, where the dynamic image series is converted to a single image with each voxel describing a given quantitative physiological parameter. A simple way to think about this concept is that the count density variation in each of the dynamic image voxels represents a TAC which is analysed either on its own or in combination with other voxels characterized by the same dynamic behaviour and, therefore, kinetic model. In this specific case, therefore, it is not necessary to extract ROIs based TACs.