The idea to apply concepts, devices and software that were developed in the context of HEP experiments to medical imaging is rather natural and inspiring. The similarities become more evident when it comes to the detection of high energetic gamma photons. In a PET (Positron Emission Tomography) or a SPECT (Single Photon Computed Tomography) scanner one can find many components that are also used in physics experiments: fast and luminous scintillators, high-sensitivity photodetectors, fast electronics and last but not least powerful software to simulate the devices in the development phase.

The PET imaging technique is complementary to the well-known radiography. While radiography is able to provide very detailed morphological information of our body (or the body of an animal), PET can tell us a lot about its functions but with a coarser resolution.

How does PET work?  PET exploits the fact that matter and antimatter, when coming close, annihilate to pure energy. More specifically, in the case of PET, an electron and a positron annihilate creating two gammas which are emitted back-to-back. The positron originates from the radioactive decay of a nucleus (e.g. Fluor-18) which was attached to a biologically relevant molecule injected into the body. The electron is one of the myriad which we have naturally in our body.  The most frequently used molecule is ¹⁸F-FDG, a so-called radiopharmaceutical, which is essentially a glucose (sugar) molecule of which a small part has been replaced by the radioactive Fluor isotope. Once injected in the body, the ¹⁸F-labelled sugar will be distributed in the body according to the specific energy needs of the organs (metabolism) but may also indicate the presence of unwanted energy consumers like tumors, or hint at abnormal changes in the brain which may be signs of serious diseases like Alzheimer.

AX-PET is measuring the gammas from a so-called phantom, which consists of sets of small channels in a cylndrical plastic container. The channels are filled with ¹⁸F in aqueous solution.

From the detector point of view, a PET scanner is a fast ring calorimeter with good spatial resolution, which surrounds the object to be imaged; a geometry which looks quite a bit like e.g. the CMS electromagnetic calorimeter.  It needs to detect the two gammas from the positron annihilation in a short coincidence window in order to ensure that they belong to the annihilation of the same pair and not coming from two different ones. As the involved energies are much smaller than in a HEP experiment, the scintillation crystals of a PET detector have typically millimeter dimensions. A single detected gamma pair tells us unfortunately only little about the position of the annihilation. We only know that the annihilation happened somewhere on the line connecting the two detection points. In order to reconstruct the 3-dimensional distribution of the labeled sugar in the body, we need to record millions of such lines under all possible angles. A method known as tomographic reconstruction derives from this data sample PET images which the physicians can interpret.      

The AX-PET collaboration (www.cern.ch/ax-pet), formed in 2008 and based at CERN, set out to demonstrate a new geometrical concept of a PET scanner. Even though the PET principle was invented more than 50 years ago, R&D on PET instrumentation is still very active. The main goals are better spatial resolution, higher efficiency, better background suppression, and last but not least compatibility with strong magnetic fields in order to combine PET and Magnetic Resonance Imaging (MRI) in a single scanner. While the scintillation crystals in a classical PET scanner are quite short and indeed arranged like in the CMS ECAL, namely radially and pointing towards the ring centre, we proposed a completely different geometry: very long and axially oriented crystals. Instead of reading the crystals in blocks by bulky photomultiplier tubes (still the standard in PET), we developed an individual crystal readout by novel compact solid state photodetectors which were developed in the last two decades, mainly in the HEP environment, and are now known as SiPMs.

Our axial geometry promised uniform resolution over the full field of view and full scalability. This means that the sensitivity of a scanner can be improved by adding extra crystal layers without any negative impact on its spatial resolution. However, everything comes at a price: we also needed to measure the photon interaction point in the axial direction. The idea to use the so-called wavelength shifters strips, known since decades in calorimetry, allowed achieving this with millimetre precision.

But how did it come to the formation of the AX-PET collaboration? The idea to demonstrate an axial PET geometry was born already a couple of years before the collaboration was formed. In the late nineties some of the later AX-PET members were involved in a quite ambitious undertaking at CERN. We were developing so-called Hybrid Photon Detectors (HPD) as one option for reading out the RICH detectors of the LHCb experiment. HPDs are segmented photodetectors, which would also permit detecting light from a matrix of crystals, as used e.g. in PET. As we had developed the full production facility for HPDs at CERN, we could also adapt their shape to the rectangular footprint of a crystal matrix and build what we called PET-HPDs. We foresaw to  derive the axial coordinate along the 10 cm long crystals from light sharing at the two ends of the crystals. It turned however out that the light absorption in the crystals was quite low (the crystals were too good!) and it was difficult to achieve axial resolutions  well below a centimetre. 

The idea to build an axial PET with dual side HPD readout seemed attractive to whomever we talked and we were encouraged to apply for a patent for this technology. The disappointment was deep when we had to learn more than two years after the application, that a Japanese photodetector company had already patented a very similar idea. We had apparently done only a reinvention, but it also meant that the idea couldn’t be completely stupid, either.

In 2008, when we set up the AX-PET project like a HEP experiment, several new teams joined and brought new ideas, resources and in particular also competence in medical imaging. This allowed developing in parallel to the detector hardware also detailed simulation and reconstruction software which was optimized for the uncommon axial geometry. It took us a bit more than two years to build a fully operational demonstrator scanner which for cost reasons consists only of two detector heads at 150 mm distance. The full ring coverage is emulated by mounting the object to be imaged on a rotating table in the centre between the two modules. We tested it for the first time in spring 2010 at a dedicated PET lab at the ETH Zurich. At this early stage we used phantoms, i.e. geometrical objects filled with ¹⁸F in aqueous solution. The handling and use of such open radioactive sources are not permitted at CERN for safety reasons and we were more than happy when we learned about the company AAA, in a certain sense a spin-off from CERN, located just a stone’s throw away from CERN in Saint Genis. AAA is specialized in the production of FDG and delivers hospitals within a radius of several hundred kilometres. They allowed us to install our demonstrator for two measurement campaigns in their premises and filled our phantoms with ¹⁸F solutions. With a well understood and fully characterized AX-PET scanner, we performed our fourth and last measurements series, this time imaging mice and rats, again at the PET lab of the ETH Zurich.

Reconstructed PET image of the phantom. The channels' diameters range from 1.2 to 4 mm. Channels below 2 mm are not clearly resolved.

3D rendered image of a rat skeleton produced with AX-PET. A small amount of ¹⁸F in aqueous solution was injected in a rat. Unlike FDG,  ¹⁸F accumulates in the bones. In this case, a PET image gives no functional information but rather looks like a CT X-ray image.

The results of all these measurements have confirmed our high expectations. The scanner showed the predicted performance and led to very clear reconstructed PET images. The expected benefits like constant resolution over the complete field of view and sensitivity increase from recovering Compton interactions could be clearly demonstrated. Recently we also demonstrated that AX-PET, despite the use of long crystals, can be enhanced by a high resolution Time-of-flight (TOF) component, which promises further improvement in image contrast. The use of SiPM photodetectors makes the concept also inherently MRI compatible.

All AX-PET results were presented at numerous conferences and published in more than 20 publications and conference proceedings. The topic was ideal for several theses and post-doc projects, three of them supported by European Marie Curie funds.

During a collaboration meeting in Italy, we were enjoying a cake with the AX-PET logo. The picture doesn’t show all AX-PET members. A few missed this pleasure.

Even though our axial concept is by now well known in the PET instrumentation community and several other university groups have embarked on axial PET studies, we have not yet managed to attract a medical instrumentation company to build an axial full ring scanner. It may be that our geometrical concept is too diagonal to the traditional PET concepts. The intrinsic advantages of the concept, which entail also some additional complications, have apparently not yet convinced the few big players in the market. Even if this last step may fail to appear, we consider the AX-PET project a great success, an enjoyable undertaking for the collaboration, an ideal topic for the training of young people and, for most of us, a very nice side project to our main commitments in one of the big HEP experiments.

During the 2013 CERN Open Days, the AX-PET prototype has been one of the highlights of the busy PH-DT visit point. Visitors had the chance to discuss with AX-PET collaborators some of the advantages of the new technique and its potential applications.

The AX-PET collaboration

Istituto Nazionale di Fisica Nucleare (INFN) , Sezione di Bari

Università and INFN Cagliari, Cagliari, Italy

Ohio State University (OSU), USA

European Organization for Nuclear Research (CERN), Switzerland

University of Michigan, USA

University of Oslo, Norway

Tampere University of Technology, Tampere, TRE 33720 Finalnd

Instituto de Fisica Corpuscular (IFIC) University of Valencia, Spain

Eidgenössische Technische Hochschule (ETH) Zurich, Switzerland

References

P. Beltrame et al., Nucl. Inst. & Meth. A., vol. 654 no 1 (2011) 546-559.

J. Gillam et al., Phys. Med. Biol. 58 (2013) 2377-2394.