EP0846275A1

EP0846275A1 - An imaging apparatus

Info

Publication number: EP0846275A1
Application number: EP96928600A
Authority: EP
Inventors: Philip Palin Dendy; Edwin Philip Wraight; Steven David Bloomer
Original assignee: BTG International Ltd; British Technology Group Ltd
Current assignee: BTG International Ltd
Priority date: 1995-08-25
Filing date: 1996-08-27
Publication date: 1998-06-10
Also published as: WO1997008569A1; IL123457A0; GB9517491D0; CA2230470A1; JPH11511551A; AU6832096A

Abstract

The invention relates to an imaging apparatus (1) incorporating a dual-headed gamma camera modified for use with Positron Emission Tomography. Gamma cameras have been used in positron emitting tomography (PET). Collimators on each of the gamma cameras have reduced the amount of signal which the gamma cameras were able to detect. The present invention provides a modified dual-headed gamma camera which has the ability to image single photon emitting events without any limitation as to solid angle. The invention is more sensitive than existing systems. Use of fast digital signal processing (DSP) techniques and deconvolution using a point response function provide an enhanced image of an object.

Description

AN IMAGING APPARATUS

The present invention relates to an imaging apparatus, and more particularly to an imaging apparatus which enables a dual-headed gamma camera to be used for routine diagnostic imaging of positron-emitting radionuclides, as well as single-photon-emitting radionuclides. Such a device is hereinafter referred to as a modified dual-headed gamma camera.

A modified dual-headed gamma camera is used to image positron-emitting radiopharmaceuticals in routine clinical applications. The present invention describes a modified dual-head gamma camera which is capable of retaining the ability to image single-photon-emitting radiopharmaceuticals. The invention thus enables opposing gamma camera detectors to operate in coincidence mode, thereby obviating the need for collimators. The subsequent improvement in sensitivity over a collimated system is exploited by use of fast digital processing techniques to handle efficiently the relatively high rate of single events, and by the application of a fully three-dimensional image reconstruction technique to ensure efficient use of acquired data. The invention may be incoφorated as a switchable unit in future gamma camera system designs, or it may be retro-fitted as a modular add-on to existing designs.

An example of an existing gamma camera imaging system is described in granted United States Patent Number US-A-4057727 (Muehllehner et al). The arrangement described comprises a positron imaging system having two opposed Anger cameras, each camera having an approximately one inch (0.0254m) thick unity layer of scintillation crystal. However, the arrangement suffered from the drawback that, because such a thick layer of scintillation crystal was present, there was poor event separation which gave rise to poor spatial resolution. Because of the poor event separation the system also had a low count rate. Additionally it was not possible to use all events for image reconstruction.

Radionuclide imaging is a means by which the physiology of a patient can be studied and is therefore used as a tool in patient diagnosis and treatment management. The basis ofthe technique is to introduce a radioactive substance into the patient after attaching it chemically to a physiologically important pharmaceutical, and to use gamma ray emissions from this radiopharmaceutical to construct images of its distribution at subsequent times under certain controlled conditions. The clinical information obtained from the test depends on the biochemistry of the radiopharmaceutical, and the types of radiopharmaceutical available are often limited by the chemical properties of the radioactive element to be attached.

The commonest form of radionuclide imaging is to study gamma ray emissions from radionuclides that emit single gamma ray photons of energies in the range 50-400keV. The tomographic study of such emitters is usually termed SPET or SPECT, though static and non-tomographic whole body scans are common. The equipment used to image these emissions is the gamma camera, which is a position sensitive gamma ray detector. A gamma camera detects gamma rays using a scintillation material (usually a single crystal of thallium-doped sodium iodide). Gamma rays incident on the scintillation material convert some of their energy into optical light which is detected by a number of photosensitive detectors (usually photomultiplier tubes optically coupled to the scintillator by a lightguide). The detector obtains the directional information of each gamma ray by means of an absorbing collimator, which limits the solid angle of acceptance to a narrow range around a particular set of directions.

It is possible to use several planar images obtained at different orientations around the patient to reconstruct a 3-dimensional (tomographic) image ofthe radiopharmaceutical distribution.

Another form of radionuclide imaging is positron emission tomography (PET). In this case, the radioactive element emits a positron when it decays, which then collides with a nearby electron in the patient. The positron and electron undergo mutual annihilation resulting in the emission of two 51 IkeV gamma rays in near-opposite directions. The near- collinear property of these photons is useful in determining the location of the emission point in the patient and collimation is not necessary if both gamma rays are detected. In fact, because the energy of the gamma rays is so much higher than those encountered in SPET, gamma cameras are not well-suited to detecting them. PET radionuclides are therefore usually imaged with a dedicated PET imager, which consists of several rings of position-sensitive scintillator/photomultiplier detectors operating in coincidence mode (i.e. without a collimator). The detectors (usually bismuth germanate, BGO, scintillators backed by position-sensitive photomultipliers) are optimised to detect 51 IkeV gamma radiation. Gamma ray events arriving at the detector rings at the same time are assumed to originate from the same annihilation event and the line between the two serves as the locus of possible emission sites within the patient. If many events are recorded, a tomographic image ofthe radiopharmaceutical distribution may be reconstructed.

SPET imaging with gamma-cameras is well established as a routine diagnostic tool and has a wide range of clinical applications in areas such as cardiology, oncology and neurology. However, the single photon emitters tend to be heavy elements (e.g. technetium ^99mTc) and the range of pharmaceuticals to which these can be attached is limited. Positron emitters, on the other hand, tend to be lighter elements (e.g. fluorine ^l8F) and are more readily attached to biologically important chemicals. For this reason, PET has been developed over the past two decades into a highly valuable research tool, providing physiological and medical imaging data not obtainable in any other way.

Recently it has become apparent that PET techniques can make a major contribution to diagnosis and patient management in routine medical practice (Schwaiger and Hutchins 1992, Broich et al. 1992 and Hawkins et al 1992). Whilst specialised centres derive great benefit from brain perfusion and functional imaging the most important areas for more general clinical application are in cardiology, demonstrating regional perfusion and myocardial viability more precisely than by any other technique, and in oncology where 18F-fluorodeoxyglucose (FDG) has been demonstrated to provide unique information on tumour staging and assessment of residual tumour viability following therapy. Despite this, PET radionuclides are not currently used as routine diagnostic imaging agents. The chief reason for this is cost. Because of their higher energy, PET gamma ray emissions are most effectively imaged by a dedicated PET imager. However, a dedicated PET imager is currently approximately six times as expensive as a state-of the art, large field of view modified dual -headed gamma camera. This is largely because PET imagers consist of several detection rings, each comprising many small, high-cost position sensitive detectors, whereas the comparatively simple modified dual-headed gamma camera consists of only two large position-sensitive detectors. In addition to this, for a dedicated PET imager to be viable, it requires its own source of short-lived radionuclides on-site; for example a cyclotron. A complete dedicated PET facility, with imager, cyclotron, and associated buildings costs can be very expensive and accordingly, dedicated PET imaging is restricted to research use in specialist centres. Routine diagnostic scans are simply too

-J- expensive.

There have been several attempts to develop a low-cost alternative to the dedicated PET scanner. However, none of these is suitable for routine diagnostic imaging, either because the cost remains prohibitively high, or because the count rate performance or sensitivity is too low for practical routine use. If the count rate performance is poor, scans simply take too long. If the sensitivity is too low, a high amount of radioactivity is required to complete an acquisition in a reasonable time, and the radiation dose to the patient may be too high for routine or repeated use.

According to the present invention there is provided an imaging apparatus having first and second gamma cameras, arranged such that the gamma cameras are substantially diametrically opposed, means for detecting an event, processing means for counting events and for manipulating data obtained from events so as to construct an image of an object, characterised in that the solid angle ofthe gamma cameras is not limited by collimation. Preferably shielding vanes are provided to reduce the number of single events. A thin layer of scintillation crystal material is provided, typically of the order of thickness of 0.095 m, which is the industry standard thickness. Improved sensitivity may be obtained by removing the collimators which were present on prior art systems. This permits the modified dual-headed gamma camera to be used in coincidence mode without collimators and increases the sensitivity ofthe apparatus over existing collimated systems. Furthermore, the count rate is improved by using digital event detectors and pulse separation techniques.

In a preferred embodiment the invention is a unit that is used to enhance the performance coincidence detection and image reconstruction to a modified dual-headed gamma camera to enable it to be used effectively in routine imaging of PET radionuclides. It takes the position and energy signals from each detector and uses them to determine that an event has occurred, to distinguish overlapping events, and to perform the image reconstruction using an efficient technique. The unit may be added as a stand-alone device onto existing gamma camera designs or it may be incorporated as a switch-in integral module in future designs. In either case, the invention can be easily disabled to allow the system to operate in single-photon mode.

Shielding vanes may be necessary to reduce the number of single events reaching the detectors. Detector tuning and calibration, and gantry operation may be controlled by the host system.

In addition to these preferred features of the invention, Digital Signal Processing

(DSP) means may be provided for enhancing the speed of processing data. This allows the high singles count rate, which is a direct consequence of the thinner crystal layer, to be handled. Preferably Fast Fourier Transformation is applied to the data, using said DSP means.

Sensitivity and count rate performance may be further improved by application of a fully 3 -dimensional reconstruction technique because this uses approximately all ofthe time coincidence events; unlike previous systems, which have rejected many of these counts.

The invention described in this Application describes a viable low-cost alternative to the dedicated PET imager that addresses system performance and is therefore the first device that will facilitate the use of PET radionuclides in routine medical practice. In a preferred embodiment the imaging apparatus allows single-photon, as well as positron- emission tomographic, whole body and planar scans to be carried out on the same system. The imaging apparatus' performance cannot compete with a dedicated PET scanner, and it can only image longer-lived radionuclides, such as ¹⁸F, but can produce clinically useful results and with its low cost it is the only practical method for routine imaging of PET radiopharmaceuticals.

The main low-cost alternatives to the dedicated PET imager are discussed below:

The first is a Multiple Wire Proportional Chamber (MWPC). The multiple wire proportional chamber is a form of position sensitive ionisation chamber comprising a large area gas filled chamber containing an array of fine wires. In a PET system, two opposing MWPC s are operated in coincidence. The chief advantage of a MWPC system is its low cost. MWPCs require a solid photon-electron converter within the chamber to make them suitable for detecting 51 IkeV gamma rays. Ott (1993) reports that most MWPC detectors achieve a single-photon detection efficiency up to 20% at 51 IkeV. While this is better than gamma-camera efficiency, the detectors are very susceptible to scattered radiation and have no energy discrimination ability. They therefore suffer a great deal from single event count rates, and accidental coincidences. As a result of this MWPC systems actually have low count rate capability: the MUPPET system has a maximum data rate under 3kcps. The HIDAC system (Townsend et al. 1987) has a maximum count rate capability of around 4kcps. This low data-rate is the main reason why the MWPC systems have not become commercial products. Ott notes that the improved efficiency of a hybrid detector consisting of Barium-

Fluoride scintillator in conjunction with a MWPC filled with UV -photon sensitive gas (TMAE) may provide an effective, low-cost alternative to BGO systems.

The second low cost alternative is a Cost-Reduced PET Imager. The low data rate of MWPC detectors has discouraged some researchers from this approach. A group in Geneva are working on a project to develop a rotating PET scanner using BGO block detectors, with the help of Siemens, (Townsend et al. 1993). This system is a cut-down version of a static dedicated PET imager, and comprises two rotating, opposing sections of a PET detector operating in coincidence. This system has the advantage that it uses detectors optimised for 51 IkeV. However, it cannot image single photon emitting radionuclides, and, although it is doubtless less expensive than a full dedicated PET imager, it is estimated to be more expensive than the present invention.

The relative low cost of gamma camera systems make them an attractive possible alternative for low-cost and hence routine PET imaging. However, the gamma camera is optimised to image at energies around 140keV and cannot operate at 51 IkeV without modification. If the modification is such that the original system set-up can be reverted to easily, these systems have the advantage that they have the dual capability of single-photon (SPET) work and routine PET imaging. Several attempts have been made at modifying a gamma camera to image PET radionuclides. Some are briefly described below.

The most obvious method of modifying a gamma camera to image PET radionuclides is to use a special collimator optimised for 51 IkeV radiation, and to operate the detectors under a different set of control parameters and with a simple software modification to enable them to accept the corresponding signal sizes. This method has been used successfully to image PET radionuclides by several groups (Van Lingen et al. 1992. Kalff et al. 1994) and nuclear medicine (SPET) system manufacturers are now offering these collimators as product or work in progress with varying amounts of success (Britten et al. 1995. Clarke et al. 1995). The first problem with imaging 51 IkeV radiation with a collimated gamma camera is that the probability of a photopeak interaction in the crystal is much lower than at 140keV. Production gamma cameras usually have a crystal thickness of 9.5mm, at which the photopeak interaction efficiency for 51 IkeV radiation is only 11.8%, compared with 90% at 140keV (Anger 1964). This, combined with the inherently poor efficiency of the collimated method of image formation, results in a poor system sensitivity. A typical level of administered activity needed to obtain an FDG scan in a reasonable time with a collimated gamma camera system is around 370MBq, resulting in a total radiation dose to the patient of approximately lOmSv. Such levels are too high to be recommended for routine use in many clinical situations. A second problem comes from the relatively low attenuation coefficient for 511 keV gamma rays, compared with those for gamma rays at the energies for which the gamma camera is designed as can be seen from the graph in Figure 3.

Consequently, the amount of high-atomic number material (usually lead) needed to bring about the same level of absorption as at 140keV is greatly increased, and collimators designed for 51 IkeV therefore tend to be heavy if they are to be effective. Gamma camera systems rotate and need to be balanced, and this imposes weight restrictions on the collimators that can be attached. Such weight restrictions can lead to compromises in the collimator design and subsequently their imaging performance can suffer. The commonest problem is a high level of penetrating radiation, which reduces contrast in the final image. Such problems have been seen in practice to preclude certain types of clinical study (Clarke et al. 1995).

The present invention has a sensitivity which is many times that ofthe collimated gamma camera and therefore can be used with lower activities and consequently lower radiation doses to the patient. It is therefore suited to routine diagnostic imaging. It does not suffer from weight restriction problems, because it does not need heavy collimators, and the spatial resolution is likely to be better than collimated systems because the present invention does not rely on collimator hole geometry.

An alternative to the collimated approach is to make use of the dual-gamma nature of PET radionuclide emissions in a similar way to the dedicated PET scanner by operating two opposing gamma cameras in coincidence detection mode. Several groups have attempted this, with poor results (Paans. Vaalburg and Woldring 1985, Sandell et al. 1992). The system developed by Paans, Vaalburg and Woldring could also converted back for SPET use by means of a switch. The overriding problem with the systems is their poor count rate capability. The reason for failure is apparently two fold. The first is that the systems used in these early trials had poor pulse separation electronics. The second is that only a small fraction of recorded events were used in the final image reconstruction.

The count-rate limiting factor is that the system must handle a large proportion of events where only one of the two gamma rays has been detected. With a photopeak detection efficiency of 12%, and with typical source geometries a true coincidence count rate of lOkc/s means that each detector is handling a singles count rate of around 200kc/s. At this singles rate, approximately 18% of events from a single detector overlap one another to some extent and the processing electronics must be able to handle this without loss of counts and with minimal loss of resolution. The systems used in early experiments were unable to achieve this.

The main advantage of the present invention over the collimated system, is its efficiency, because although the chances of seeing a twin gamma ray event is only 12% of that of seeing one of the two gamma rays (as in the collimated case), there are no collimators and thus the solid angle is increased by around a factor of 1000. However, this solid angle must be exploited if the system is to be of practical use. Early experiments did not exploit the increased solid angle available because of problems in the reconstruction process. Only when an impracticably small reconstruction volume was adopted did the early attempts yield clinically useful count rates.

The present invention adopts a novel approach to overcome the aforementioned problems. It uses fast digital hardware to improve the pulse separation and a modified data- efficient reconstruction technique to include 100% of recorded data in the image reconstruction. The resulting device has a count rate capability which is suitable for most routine clinical applications with the longer lived PET radiopharmaceuticals.

An embodiment ofthe present invention will now be described, by way of example only, and with reference to the Figures in which:

Figure 1 shows a cross-sectional schematic of a typical gamma camera; Figure 2 shows a Schematic view of a dedicated PET imager in operation;

Figure 3 shows a graph of linear attenuation coefficient against energy; Figure 4 shows a schematic overview ofthe invention in use with a modified dual- headed gamma camera;

Figure 5 shows a functional schematic view of PET unit;

Figure 6 illustrates the formation of a coincidence trigger, and pulse pileup separation;

Figure 7 shows diagrammatically that the point source response function is not spatially invariant for a detector pair working in coincidence, in which shaded areas represent the point response functions for a point near the edge ofthe FOV (a) compared with that near the centre ofthe FOV (b); and Figure 8 shows an axial slice of shell phantom containing two point sources: a) shows a simulated back-projection b) shows a forward projection from initial estimate c) shows a final deconvolved image d) shows an original simulated emitter distribution.

Reference will also be made to a TABLE of results which summarises the count rate capability study for 60 x 46 cm detectors in a worst case scenario. Figure 1 shows diagrammatically a gamma camera comprising a plurality of photomultiplier tubes 2, a lightguide 3, a scintillator 4 and a collimator 5. Figure 2 is a diagrammatical view of a patient 6 in a position sensitive detector ring 7 in a PET imager.

Referring now to Figure 4 which shows how energy and position signals from detectors 10 of a gamma camera 11 are fed into a processing unit 12. The signals represent raw analogue energy and position signals (E, X+, X-, Y+, Y-) from an Anger position matrix 14. The following description therefore relates to such signals. However, it is understood that the application ofthe invention need not necessarily be restricted to them as the general principles may well apply to alternative applications.

The signals have the pulse-shaping particular to the make of gamma camera 11, but unless pole-zero cancellation has been implemented the signals will probably exhibit the characteristic time constant ofthe scintillator. This is 230ns for NaΙ(Tl) .

Referring now to Figure 5, the event trigger is carried out using the energy sum signal from each detector A and B (channel A and B respectively). Suitably conditioned signals (i.e. with a suitable risetime) are fast-digitised using a flash converter running at a clock speed of 40MHz or more. The resolution ofthe flash converter needs to be such that final integrated signal resolution is not unduly impaired by over-course digitisation. Adjacent sets of samples are tested on each clock cycle to see if there is a gradient greater than a predetermined threshold until both channels show a positive result (e.g. trigger point Q as shown in Figure 6), at which point a coincidence event is detected. The duration ofthe coincidence time window determines the number of random coincidences recorded (i.e. events which are causally unconnected). For a 100ns coincidence window, and with a singles rate in each detector of 200kc/s, the fraction of all measured coincidences which are random will be around 20% for typical detector and patient geometries. For this reason, the coincidence window needs to be as small as possible but at the same time keeping a sufficient signal-to-noise ratio. Digitisation at the speeds suggested above facilitate the use of coincidence windows of less than 100ns.

The event stream from each detector suffers from overlapping events. At a singles count rate of 200kc/s, 18% ofthe events in each stream overlap one another at least once. It is therefore necessary to take this into account when integrating under the position and energy signals. This can be done by using the single channel trigger to determine when each pulse starts. By integrating the signals into accumulators and storing the sums at the onset of each trigger it is possible to separate the overlapping events by using knowledge ofthe characteristic decay ofthe signals. Details of pulse integration are shown in Figure 5.

Figure 6 depicts one possible sequence of overlapping events on energy channels A and B. Only two events are coincident. P, Q and R are the first three trigger events detected by the gradient test. At trigger P (originating from channel A), channels A and B start being integrated into a separate accumulator, until trigger Q occurs. The integrated sums for channel A and B are stored temporarily. Note that trigger Q is also a coincidence trigger, because it is detected in both channels A and B. The accumulators for channels A and B are then cleared and start integrating again, until the next trigger (or a suitable time longer than the typical signal envelope). In this example a third trigger, R, occurs because of a lone pulse on channel B. At this point, the accumulator contents for the integral Q-R are stored temporarily and this trigger/integration process continues.

At this point, because there has been a coincident event, the system attempts to separate the events on channels A and B. For channel A, the coincident event is "sitting on the back" of a preceding event. The stored integral P-Q for that channel may be used to determine the residual tail of the preceding event upon which the event of interest rests. This can be done by multiplying the integral by a value stored in a lookup table containing the characteristic decay of a signal. The residual tail is then subtracted from the integral stored for Q-R, which is also used to determine the residual tail which was lost because of the trigger from a following event on channel B. This residual tail is than added to the result to give the value ofthe coincident event energy signal for channel A. The energy value for channel B is arrived at in exactly the same way. Applying the same algorithm has the same effect.

The position signals for channels A and B are separated in a similar fashion, except that they take their trigger points levels from the energy channels. In this way it is possible to detect coincident events and handle pulse pileup without loss of events or undue loss in resolution. Baseline shifts caused by the build up of the delayed component of the scintillator output can also be corrected digitally during the integration process.

The arithmetical processes used in this determination is possible in real-time using digital signal processors (DSPs).

Signal normalisation and position determination are then be carried out for each detector as shown in Figure 5. This can be carried out digitally in the generally accepted fashion, whereby the X coordinate in the field of view is given by:

(scaling factor) x (X+ - X-) / (X+ + X-)

The denominator serves to remove energy dependence in the result. Some systems may use the total position sum (X+ + Y+ + X- + Y-) here. The Y coordinate is achieved in a similar fashion.

The spatial coordinates are then used to determine the energy and spatial corrections which need to be applied to the energy and position ofthe events in each detector. This can also be done digitally using lookup tables. Energy and spatial errors are an unavoidable feature of gamma camera design and there are many techniques to determine and correct for them.

Obtaining spatial corrections at 51 IkeV is made difficult by the same problem that besets the generation of spatial corrections at the higher energies in SPET work (e.g. at 364keV for ^13,I). A line or spot phantom does not absorb enough radiation to produce an effective correction map. The most probable solution is to tune the system for 51 IkeV such that the signal sizes (and hence spatial distortions) are similar to those for ^99mTc in the single-photon mode, where spatial corrections are possible, and adopt the spatial corrections obtained with ^99mTc in single-photon mode for use at 51 IkeV.

As a first step towards scatter rejection, an energy discrimination level should be adopted for each coincidence event. Both single events making up a coincidence must be included in the window for the event to be accepted. The choice ofthe position and width of the energy window will depend on the energy resolution of the system and on the usefulness of Compton-scattered radiation. Some radiation will Compton-scatter within the detector such that only part ofthe energy of a gamma ray is deposited in the scintillator 4, others will have scattered in the patient and therefore lost useful spatial information, some will have scattered in both patient 6 and detector 7. However, many of these events may possess correct spatial information because they have not been scattered in the patient. Once the two corrected positions have been determined for a coincidence event, and if the corrected energy lies within the energy discrimination window, the orientation ofthe line of response (LOR) between the two coordinates can be calculated. For this calculation the detector orientation and separation must be known and these values therefore need to be obtained from the host system at intervals during the acquisition. Once the LOR has been calculated, the event is characterised by the energy of each gamma ray and a direction through space and can be used in the image reconstruction process. The reconstruction process is described in the next section. The event information may either be stored in sequential 'list' mode for subsequent processing, or used in real¬ time back-projection if the processing equipment has sufficient capability. Data Efficient Image Reconstruction Technique.

If the reconstruction cannot use all of the true coincidence data the effective maximum count rate performance is likely to be reduced. A maximum count rate of 1 -2kc/s might be expected for typical reconstruction scenarios (e.g. Townsend et al. 1987) in which 80% of all coincident events are discarded. Early attempts at getting gamma cameras to work in coincidence failed partly because of this reason. This section explains the reasons why such a high proportion of events were rejected and explores some alternative, more data-efficient, reconstruction techniques that have been developed. One of these methods has been adapted for use with this device.

Fully three-dimensional (3D) image reconstruction uses gamma-ray events with lines-of-flight having components in the axial direction as well as those travelling in the transaxial plane. Single-photon (SPECT) systems employing parallel-hole collimators and typical multi-ring PET scanners with inter-ring septa use only those gamma-rays travelling at angles close to a transaxial plane and are inherently inefficient for this reason.

The problem of 3-D reconstruction has been considered for some years (see e.g.

Jeavons 1980, Colsher 1980, Daube-Witherspoon and Muehllehner 1987) and most attempts at its solution comprise a process where each individual event, or a set of filtered

2D projections, is back-projected through a 3D volume, followed by the deconvolution of the system point source response function from the backprojected image in Fourier space.

The 3D back-projected image g(x) can be thought of as the convolution of the radionuclide distribution f(x) with the backprojected image of a point source measured with the same system, h(x,_x'), where h(x, x') denotes the contribution to the image at position x due to a point source positioned at x' :

g(x) = /f(x)h(x,x')dx (1)

If the point response function h(x, x') is known, and g(x) is the acquired, back¬ projected image ofthe radionuclide distribution, the true radionuclide distribution, f(x). can be isolated by deconvolution via use of the discrete Fast Fourier Transform. Analytic expressions for the point response function (or the appropriate filter for the deconvolution) have been derived (Colsher 1980).

The overriding caveat of this approach is that the point-response function is assumed to vary only with position relative to the point source, x', and not with the position in the backprojected image, x. In other words, the backprojected image of a point source is assumed to be the same wherever the source is positioned in the system.

Clearly this is not the case for a system using coincidence detection, because lines of response of detected events originating from near the centre ofthe detector arrangement can lie at steeper angles (θ) to the planes of the detectors than those emanating from the edge ofthe field of view as can be seen in Figure 7.

The reason why so many experimenters have rejected a large fraction of the coincident data is because they consider only those events which fall inside the maximum value of q which satisfies the condition of spatial invariance ofthe point response function, thereby allowing its deconvolution from the backprojected image. Such an approach is most suitable for small regions of interest placed closed to the centre ofthe imaged volume, where the number of events meeting the angular constraint is least. However, for practical imaging applications, a more efficient reconstruction method is essential.

An iterative approach which begins by reconstructing a small region of the backprojected image using an appropriate value of limiting q may be used. The region chosen is small enough so that the point-response function changes little within it. A surrounding region of the backprojected region is then chosen, and for which a smaller value of q is appropriate. The original image is deliberately convolved with the point- response function corresponding to the new angle, is then added to the new region and the result is deconvolved with the new point response function. The process iterates until the entire region of interest is reconstructed.

Another approach can be summarised as follows: first, a low-statistics image is reconstructed using a limited angular acceptance which meets the spatial invariance condition ofthe point response function. The point response function is the deconvolved from this back-projection (usually by prefiltering the projections) giving a first estimate of the radionuclide distribution. This distribution is then forward-projected to provide an estimate ofthe projections not measured because of the limited geometry ofthe detectors. Scaling and adding the estimated missing projections to those actually measured allows the result, which incorporates about 90% ofthe detected data, to be successfully filtered with an appropriate spatially-invariant point response function. Rather than applying a filter in 3D Fourier space, it is possible to use the analytic expression derived by Colsher (1980) to apply an appropriate filter to the 2D projections or sinograms prior to backproj ection.

The method proposed by Kinahan and Rogers (1989) and used successfully by others has been investigated further to test its suitability to this invention. However, there are some differences in the approach adopted for this study and those of others. Whereas others have organised data into 2D projections, the present method does not. In this case the forward-projection process is based on a Monte Carlo process. That is events are generated and projected randomly and individually into the solid angles which would otherwise go undetected. In this way, the sum ofthe 'measured' and forward-projected data simulate the response of a 4π detector, and 100% ofthe true coincident events can be used in the reconstruction. Although a Monte Carlo approach appears time consuming, use of sufficient processing power may make this an alternative in clinical practice. In this application, all backproj ections are deconvolved in 3D Fourier space.

Once an event has been detected, its line of response (LOR) is tracked within the reconstruction volume. The volume is split into a voxel matrix. Note that the reconstruction volume can be any size. It can even exceed the physical limits ofthe system if necessary. Sufficient memory is allocated dynamically by the software to accommodate the contents ofthe matrix. Each point is represented by a floating point number (though long integers could be used to improve speed). The LOR is the straight line between the points of intersection ofthe gamma ray trajectories with the detector planes. The point of emission ofthe original positron is assumed to lie along this LOR.

In a real system it may not lie along the LOR. The positron may have moved some way before annihilating, or the two gamma rays may not be quite collinear due to the motion ofthe centre of mass ofthe positron-electron pair. Or this may occur because the two detected gamma rays forming the coincidence are causally unconnected. Nevertheless, in a real system the point of origin is assumed to lie on the LOR, and the fact it may not is a contributing factor to noise.

The first step in the backproj ection process is the calculation of the points of intersection ofthe LOR with the bounding box. This helps to define the range of voxels through which the LOR has to be tracked. The point of intersection ofthe LOR with each of the inter-voxel planes it crosses is calculated and the distance of this intersection from the point that the LOR enters the bounding box is stored along with the index ofthe plane it is crossing. If the LOR is parallel to a voxel side, it will have a length in that voxel of 1.0. If it passes through diametrically opposite corners of a voxel it will have a length in that voxel of 1.732. These distances are then sorted sequentially as follows. The shortest distance stored is the length ofthe LOR lying within the first voxel. This length is added to the image memory at the address of that voxel. The next largest distance minus the previous distance is the length ofthe LOR passing through the adjacent voxel, and is added to the image memory at that voxel's address. This cycle is repeated until the entire LOR has been accounted for in the reconstruction volume. By adopting long integer rather than floating point arithmetic, the process is optimised for speed. The backprojected image represents the distribution of emission points smeared with the backprojected image of a point source (the point response function). For a detector with limited geometry, the point response function will vary with position, and the distribution of emission points for an extended source cannot be retrieved simply by deconvolving the backprojected image with some nominal point response without generating artefacts in the result.

As discussed above, the backprojected image of events meeting the condition of point response function invariance, can yield a low-statistics estimate ofthe distribution of emission sites. This low-statistics image can then be used to forward project generated events into those solid angles ofthe system where events are not normally detected. When this forward-projected image is scaled and added to the true backprojected image, the result can be deconvolved with the point response function obtained for a 4π detector.

The forward projection process works in the program as follows. First, a special patient scan is carried out where the angle of acceptance is limited to the maximum that can be achieved for a particular imaging volume and detector size. In practice, this could form a subset ofthe events detected in normal acquisition, and a copy of these could be stored separately during the acquisition: a separate acquisition may not be required. These θ- limited events are backprojected as described above, and the resultant image is then deconvolved with the corresponding point response function. This response function is determined empirically by generating events under the same conditions from a point-source placed at the centre ofthe imaged volume. A frequency filter, such as a Harming window, (see e.g. Colsher 1980) with the cut-off set to the Nyquist frequency, is used to minimise the effect of noise in the deconvolution process, as described in detail below. The deconvolved low-statistics image is then used as the generator for artificial, forward- projected events. The low-statistics generating image is prepared for event generation by first setting voxels with contents lower than a suitable threshold. The remaining voxels are then scaled so that their total equals the number of events to be generated. The forward-projection process then steps through the non-zero voxels and uses their centres as emission points. Event direction cosines are generated as before. This time, if the event does not meet the criterion of detection, its LOR is tracked through a forward-projection volume in exactly the same way as in the backprojecfion case. In this case, no limit on θ is applied. The generating image voxel content corresponding to the emission point is then decremented. The process repeats until that generating image voxel is empty, in which case the program moves on to the next non-zero voxel. The program proceeds in this manner until the generating image is depleted. Once the forward-projection is complete, it is scaled and added to the original, unlimited-θ backprojected image of the phantom. The scaling factor used is determined from the ratio of the forward-projected events that do not meet the 'not-detected' to the number of events detected in the original back-projection.

Finally, the summed image is deconvolved with a point response function empirically determined for a 4π detector to give the final result.

The deconvolution ofthe empirically-determined point response functions from the backprojected data is carried out in the Fourier domain. For this purpose, the discrete 3- dimensional Fast Fourier Transform (FFT) is used. The Nyquist critical frequency is set by the imaging matrix. A filter such as the Harming filter (as used by Colsher 1980), with the cut-off frequency set to the Nyquist frequency is used prior to carrying out the final inverse transform to reduce the effect of noise on the deconvolution.

Note that the convolution theorem assumes that the response function and main data structure are periodic in each direction. This is clearly not really the case. The point response function is wrapped around prior to deconvolution, but the data is left intact. When convolving data in such circumstances, it is usually sensible to zero-pad one end of the data area in each dimension to avoid the effect of overlapping the convolution of one side ofthe data area with data from the opposite side. Deconvolving data which is non-zero at the image edges with a point response function, which is very broad compared with the size of the image, similarly gives erroneous results whether the data is "zero-padded" or not. If the data is zero-padded, the backprojected image (which is really a convolution) then effectively has a shaφ edge which gives artefacts when deconvolved. If there is zero-padding of the data, it is necessary to deconvolve one edge of the data using information from the opposite edge. The only way to alleviate this problem is to ensure that the important region of the data is at least a point response function's width inside the image boundary. This is an important factor in the implementation of this technique. An important potential use of a gamma-camera PET system is oncological whole- body scanning and static imaging. In whole body or static planar imaging, detectors do not rotate about the patient and therefore there is further restriction to the directions ofthe lines of response (LORs) available for use in image reconstruction. The static and whole body imaging capability of such a system are considered to be a particularly attractive features and may be implemented in a preferred system.

To maximise the efficiency of the system, the reconstruction algorithm uses as many of these LORs as possible. In addition, since the LORs pass through the imaging volume at different angles they carry information about the depth distribution of the emitting radiopharmaceutical so as to extract this information. In order to position a patient under the scanner, it is necessary to observe an image ofthe radiopharmaceutical distribution in real time. A solution is proposed where a subset of coincident events, which have a limited range of acceptance angles, are used directly to form a low-statistics, real-time image. This is really a form of electronic collimation. The subset of gamma ray events might be that used to determine the first estimate for the forward-projection process if the solid angle is small enough to form a suitable image.

There is now described results of an experimental trial of the invention, which is based solely on calculation and Monte Carlo software simulation.

Software simulation was written to determine the likely sensitivity and count rate performance ofthe system under different imaging situations, and to test the reconstruction technique.

For 60 x 46 cm the system is seen to be at least 65 times as sensitive as a typical modified dual-headed collimated system, with a count rate capability of over lOkcps. These results were obtained using 9.5 mm thick crystals. The following key is used in the TABLE. Rect = rectangular, Circ = circular detector. Detectors are separated by 50cm in all cases. Standard errors given are statistical. Photopeak efficiency assumed is 0.118.

Table 1:

Activity used in the system sensitivity calculations is taken to be the annihilation event rate in the phantom. The system spatial resolution is estimated at around 7.5 mm throughout the imaging volume.

The reconstruction technique described above has been tested successfully by means of a computer simulation. Figure 7 shows axial slices ofthe reconstruction of a simulated emission phantom during stages in the reconstruction process.

The static imaging process mentioned above has been tested by means of computer simulation and 2D images have been successfully obtained.

It will be appreciated that the present invention has been described by way of one embodiment only and variation may be made without departing from the scope of the invention.

Claims

1. Imaging apparatus having first and second gamma cameras, arranged such that they are substantially diametrically opposed, means for detecting an event, means for counting the events and means for manipulating data obtained from the events so as to construct an image of an object, characterised in that the solid angle ofthe gamma cameras is not limited by collimation.

2. Apparatus according to Claim 1 in which the scintillation crystal is 0.095 m thick.

3. Apparatus according to Claim 1 or 2, having shielding vanes provided to reduce the number of single events.

4. Apparatus according to Claim 3, having a digital event detector and pulse separation means.

5. Apparatus according to any preceding claim in which processing means is provided for performing a Fast Fourier Transform to data.

6. Apparatus according to any preceding claim in which means is provided for obtaining a 3-dimensional reconstruction of an object, using time coincidence events.

7. Apparatus according to any preceding claim in which it is adapted for positron emission tomography.

8. Apparatus according to any preceding claim in which the apparatus is adapted for use in single planar scans.

9. Apparatus according to any preceding claim having means for determining an energy level of a gamma ray and for assessing whether the energy level is below a determined threshold such that the energy value ofthe said gamma ray is used in an image reconstruction if the energy level meets a predetermined criteria.

10. A method of performing positron emission tomography using modified gamma cameras, characterised in that an image is obtained from a plurality of events said characterised in that image is deconvolved using a point response function so as to obtain an enhanced image.