European Nuclear Medicine Guide
Nuclear Medicine (NM) images display functional information of various organs and tissues, because each pixel value reflects the uptake of an administered radiopharmaceutical in a location in the patient that matches the location of the pixel in the image. Sometimes, the location and distribution can be difficult to determine, even if the uptake of activity is high, simply because there are no other uptakes aiding as landmarks for the uptake of interest. Thus, the ability to have complementing anatomical information increases the diagnostic value of the NM study. The advantage of multimodality imaging is to be able to acquire both modalities at the same time to create tomographic images that match each other both in size and in location. By using graphics, these images can be shown superimposed using different colour scales. The common method for anatomical information is Computed Tomography (CT)[130].
In the early days, the two studies were acquired separately, and this resulted in two sets of images that were not registered to each other. Different types of mathematical registration methods were suggested, but since the performance of these methods are limited, internal mismatches were expected, even if the patients were positioned on the bed as carefully as possible. Because the patients moved to different places for the two studies, the organs were also expected to misalign internally, even if the external positioning of the patient was accurate. Furthermore, transfer of data between SPECT and CT was, at that time not fully standardized and straight-forward. Hasegawa et al were one of the first researchers that designed a hybrid scintillation camera that combined SPECT with CT[131,132], and the first commercial product was released by GE Healthcare in 1999. It consisted of a low-dose one-slice CT which took about 10 minutes to acquire a full field-of-view study. The resolution was around 3.5 mm, but the possibility to perform dual acquisition aligned SPECT/CT images was a major improvement. Considering the poor resolution of a SPECT image (order 10-15 mm), the 3.5 mm spatial resolution of the low-dose HawkEyeTM CT still provided a good localization of the activity uptake. Today, SPECT/CTs are mostly equipped with high-resolution and fast diagnostic CT systems. The potential advantage is, of course, that these CT images can be used also for diagnostic purposes, and the investigation time for the patient is generally reduced, because these devices have the capability of performing two studies on the same occasion. This also is very important for the healthcare logistic.
Fusion – localization
The combination of SPECT and CT imaging has proven to be clinically useful and superior to only using SPECT by many publications, see for example[133–139]. It improves anatomic localization and adds confidence by increasing both sensitivity and specificity in diagnostic decisions resulting in an improved patient care. For example, in small regions of the spine it can be difficult to determine from a SPECT image only if an abnormality resides in the bone, or if it is located in the adjacent joints. A high-resolution CT image provides such additional information[140]. Figure 1 is an example of this improvement and shows anatomical and functional information in a fused image.
Figure 1. The images show the advantage of a fusion image (right) as compared to a planar posterior image (left) and a SPECT image (middle).
Attenuation and scatter correction
CT information can be also used for correction of attenuation in the patient and scatter in the image. These are, thus, both related to photon interactions in the patients that either absorb the photon, and thereby reducing the acquired counts, or changes the direction of the photon, with a potential mis-positioning of the event in the image. The effects from scatter and attenuation are strongly dependent on the geometry and composition of the patient, a dependence that also varies with the projection angles. These effects are usually treated separately, although they are linked to each other.
As a simple interpretation, one can understand attenuation as the reduction of the expected number of detected photons due to absorption in the patient or scattering in a new direction that makes detection impossible. To correct for this, one needs to calculate the probability for a photon to be transmitted through the patient. For this, we can use the CT images. By calculating the length of the expected line and taking into account the variation in attenuation along the line from the CT information, we can calculate a correction factor that, for each projection view, very accurately corrects the registered events. The principle is indicated in Figure 2, where the correcting factor is the reciprocal of the probability for photon #1 to travel the distance x1, where the average attenuation factor is μ1.
Scatter in the image is the result of detected photons that never should have been measured but appear in the image because of the poor energy resolution of the scintillation camera. Scatter does not produce any artefact in the images, but the locations of the scatter events are wrong. This leads to a loss of image contrast and limits the ability to obtain quantitative images. Therefore, correction should also be made for scatter. Most commercial SPECT reconstruction programs base their scatter correction methods on the use of information acquired in additional energy windows. However, since the data obtained from scatter windows generally do not represent the scatter in the primary energy window regarding amount and spatial distribution, an alternative is to model the scatter by analytical methods or from Monte-Carlo-based simulations[141].
Figure 2. Examples of two pathways for a primary photon (#1) and a scattered photon (#2) and the probabilities of escape. These probabilities are used to compensate for attenuation and scatter by including them in the modelling of attenuation and scatter in the reconstruction program.
Both methods use calculated probabilities for a scattered photon to be detected. In both cases, detailed information about the patient’s size and internal tissue distribution are essential, since photon interaction very much depends on the body composition and shape. Figure 2 shows the principles for how scatter can be modelled from CT images by calculating the probability for a photon to 1) travel the distance x2, to be 2) scattered an angle θ toward the camera, followed by 3) an escape to travel a distance x3.
Segmentation (VOI)
The spatial resolution of a SPECT image is generally much lower than a CT image by an order of a magnitude. Furthermore, image contrast is also low in areas of low uptake, which then leads to difficulties in the segmentation (i.e., the delineation either by hand or by computer algorithms) of region-of-interest (ROI) or volume-of-interest (VOI). However, if a set of CT images is properly registered to a set of SPECT images, the CT images can be used to define volume-of-interest (VOI) that can then be transferred for use on the SPECT images. For example, the geometric volumes of organs and their masses is an essential parameter when calculating the absorbed dose in radionuclide dosimetry. However, it should be recalled that spill-out (or spill-in) of counts over the outline that defines a geometric VOI will occur due to the spatial resolution of a SPECT image. Therefore, when applying a CT-based VOI on a set of SPECT images, this loss of counts due to spill-out needs to be accounted for in order to calculate the correct count density or the activity concentration if translated by a calibration factor. A method to overcome this count loss is to apply so-called recovery factors. These are often determined from experimental measurements or computer simulations of geometries with known activities and volumes and define the ratio between the total measured counts to the counts determined within a geometrically defined VOI of the source. The main limitation is that the VOI of tumours generally have arbitrary shapes, and therefore a recovery correction by a factor, obtained from a spherical source, may not completely account for the spill-out/spill-in.
Dosimetry calculations
An important field of nuclear medicine is the use of radionuclides emitting charge-particles for treatment of malignant systemic diseases. The principle is administering radionuclides or radiopharmaceuticals that accumulate close to or in the target volume, e.g. cancer cells. Damages to nearby cancer cells will occur by the interactions between the emitted charge-particles and the orbital electrons in the molecules that will cause ionizations and chemical breakdowns. Since these radiopharmaceuticals will not completely target just the main volume of interest, the radiation may also damage normal and healthy tissue. Therefore, it is important, as in external radiation therapy, to perform dosimetry to optimize the treatment with respect to the outcome and side-effects. This can be made by quantitative sets of SPECT images that display the radionuclide distribution over time, and from which mean organ absorbed doses can be calculated. In many standard treatments today, a simplified and approximate calculation scheme is used, but having access to a Monte-Carlo-based radiation transport program and a good description of the body composition (available from registered SPECT/CT study) patient-specific absorbed doses can be calculated[142,143].
As stated above, dosimetry calculations require, in principle, a procedure based on sequential SPECT/CT studies in order to determine the total absorbed dose to an organ or tumour. Then, the problem of mis-registration, due to patient movements between the time-points, appear even if each time point is acquired with a hybrid SPECT/CT study. The only way to compensate for this mis-match is to rely on mathematical registration methods, but using the SPECT data for this purpose may not be reliable due to image noise and potentially large redistributions of the radiopharmaceuticals between the time points. However, if a SPECT/CT is conducted at each time point, then a non-rigid transformation can be determined by a CT-CT image registration, and from the result of this the transformation parameters can be applied to the SPECT-SPECT images[144]. The reason why CT-CT registration is preferable, is because the CT-CT images are more similar as compared to the matching SPECT-SPECT images.
In some dosimetry applications, the calculation of the absorbed doses still relies on quantitative planar scintillation camera imaging[145]. These images are also affected by photon attenuation. The most common method to correct for attenuation is the conjugate-view method. In order to correct for photon attenuation when using this method, an image of attenuation factors is used, where each factor corrects a corresponding pixel value. If a SPECT/CT system is available, such an attenuation map can be calculated from a planar scout measurement using the x-ray unit on the SPECT/CT system[146,147]. Figure 3 shows an example of such an image. The value of a pixel in the middle image represents the probability for absorption along a line at the pixel position.
An extensive collection of SPECT/CT artefacts can be found in ref xx (SPECT/CT Atlas of quality control and image artefacts. IAEA Human Health Series no. 36, IAEA, Vienna 2019)
https://www.iaea.org/publications/13407/spect/ct-atlas-of-quality-control-and-image-artefacts (accessed 2020 05 07)
Misalignments between SPECT and CT
Although SPECT and CT studies are performed on the same occasion, the studies are sequential, and therefore mis-registrations can occur due to breathing, patient movements, scanner bed bending, or differences in acquisition protocols between SPECT and CT[148].
Figure 3. Images of a patient that undergoes 177Lu-Dotatate treatment. Left image shows an attenuation- and scatter-corrected emission image, obtained from a conjugate-view quantification procedure. Middle image shows an attenuation map, obtained from a full-body scout anterior measurement using a planar x-ray irradiation. The right image shows a fusion of the two images.
In SPECT, the acquisition time per projection is typically 15-25 seconds. This means that the patient breathes several cycles causing movement in the body due to the inhalation. The attenuation that photons experience will then be some kind of an average in the geometry. In modern SPECT/CT systems, the time required to acquire a single-slice CT is very short, and the image will in practice reflect a random time stamp in the breathing cycle, if the patient is not directed to hold their breath. Regardless, the CT image may not reflect the average attenuation over the entire breathing cycle. When using the CT images in attenuation correction or absorbed dose calculations, artefacts can be created. This is well recognized in PET, where the spatial resolution is superior to SPECT, but the same problem potentially occurs in SPECT/CT.
Because of the axial difference in the centre position between the SPECT and the CT field-of-view, a potential problem of mis-alignment occurs for bending scanner beds. To avoid attenuation problems in the couch when the camera head is located on posterior side, the couch is usually made of low-Z material. However, this might affect the stiffness, especially when imaging obese patients with CT (because the CT is often placed at the back of the scintillation camera head leading to a mis-match in the alignment between SPECT and CT images.
Beam-hardening
A general problem with CT imaging is that the images are created from an energy spectrum of x-ray photons with a substantial number of photons of lower energies. These low-energy photons have a higher likelihood to be attenuated along the line as compared to photons with higher energies. This results in an apparent lower attenuation in the centre of the CT image[149]. This effect depends on the size and composition of the patient but also on the projection view.
Image truncation of arms
Truncation of CT information, due to the systems limited field-of-view, can occur for large patients, especially if the acquisition is made with arms down. Although this effect may not be important for image fusion applications, where the main organ of interests often is located in the centre field-of-view, it can seriously affect the attenuation correction, because some parts are not accounted for certain projection views. This may result in an under-correction of the attenuation effect. This effect can be seen in the CT image in Figure 4.
Errors associated with HU conversion
The Hounsfield unit in generally defined from the following equation:
and where μ is the, so called, linear attenuation coefficient, with a value proportional to probability for interaction along the photon path. The coefficient is very much dependent on the atomic composition and on the photon energy. It should be remembered that CT values are obtained from an energy spectrum of x-ray photons with a maximum energy proportional to the voltage setting of the x-ray tube. Common voltage for a CT are 120-140 kV. In SPECT, acquisition is usually performed with a single specific photon energy. For attenuation correction and absorbed dose calculation, one needs to convert the HU numbers to attenuation coefficients. This conversion for the relevant SPECT photon energy is not trivial, and usually consist of a bi-linear regression determined from a measurement of a simplistic phantom with different materials and densities.
Streak-artefacts due to large attenuation of x-rays
In some case for very large patients and when performing a low-dose CT, streak-artefacts can occur along the directions where the pathway for the x-rays is long and with attenuation, as is the case in the pelvic region. Because the CT values are obtained from a normalized transmission measurement, large statistical variations in the normalization can occur for those directions, where the probability for transmission is very low. This effect will then translate into streak artefacts of high HUs.
Differences in spatial resolution
There is a relatively large difference in the spatial resolution between SPECT images and diagnostic CT images. This can sometimes generate artefacts in the attenuation corrected images due to the spill-out of counts into regions where no activity occurs. For example, the spill-out of counts into the lung region from surrounding tissue may generate local high absorbed doses on a voxel-level because of the low mass of a lung voxel. One method to reduce these artefacts is to degrade the spatial resolution of the CT images to a level matching the SPECT images by applying, for example, a low-pass Gaussian filter with a width that represent the spatial resolution. However, this does not compensate for large differences due to breathing.
Artefacts by contrast agents and metal implants in CT
Sometimes patient may have implants, such as, pacemakers, hip replacement, port-a-caths, nails, or stabilizers in the field-of-view that will affect the quality of the CT image. These implants are often made of some form of high-attenuating metal composition that can create streak artefacts in the reconstruction procedure. If attenuation correction is implemented in the reconstruction process, there can also be artefacts created due to very high local attenuation coefficients when reconstruction images differ very much in spatial resolution compared to CT, as is the case for SPECT. A less severe but still annoying effect is that these high CT values may have an impact on the display of the CT image. An image displayed on a computer screen often uses a finite number of colour levels evenly distributed between the minimum and the maximum CT value in the image. If there are pixels of very high values, like the ones representing the implants, then contrast of the pixel values representing the tissues will be very poor. This is illustrated in the left image of Figure 4. In this patient, there is a small metal stick placed close to the spine. The grey-level is linear, so that the details in other parts of the transversal slice are hard to visualize. In the right image, the grey-level scale is changed by a gamma correction of 0.3, meaning that a large number of available grey-levels are distributed among the lower HU values as compared to the high HU values.
Administration of iodinated contrast solutions can enhance certain information in the CT images due to the high attenuation of the x-ray, and by this create a high contrast in the CT image where the solution in present. However, these contrast solutions can subsequently affect the attenuation correction of the SPECT data, if the CT images are used for this purpose. This represents a bigger problem in PET/CT due to the large difference in photon energy between PET and CT. The results could potentially lead to an over-estimation of the activity in the areas of high concentration of the contrast agent. However, the contrast, although present in the patient, has a smaller concentration within the camera field-of-view when the SPECT study is conducted due to the difference in time between SPECT and CT, and the effect may be moderate.
The effective dose is higher when using SPECT/CT as compared to SPECT alone. The contribution from the CT exposure depends on the examined part of the body and is in the range of approximately 1-10 mSv[150]. A low-dose CT, using less milliampere-seconds (mAs), gives only an effective dose of few mSv and in many cases, such study that provides sufficient information for both localization but also for attenuation correction. However, this depends on the diagnostic purpose of the investigation. It should, however, always be remembered that the value of the diagnostic information should always exceed the increased risk of the radiation dose.
Figure 4. The figure illustrates display problem when a high-attenuating material, indicated by the arrows, is present in the patient during a CT. Left image shows a scout x-ray image where the location of the metal device is clearly seen. The middle image shows the use of a normal linear grey-level scale while the right image shows an image with a grey-level scale manipulated in a non-linear way to increase the dynamic range of levels for the voxels with values corresponding to the tissue HU ranges.