CAPÍTULO 2. ANÁLISIS Y DISEÑO DEL PROTOTIPO
2.1. requerimientos
2.2.14. Captura de datos
The earliest attempts to produce standardised instruments and methodologies that could be utilised to link local flow and metabolism were carried out by Kety and Schmidt in the 1940s (Kety and Schmidt 1945). These first attempts used freely diffusible, inert tracer substances to calculate cerebral blood flow. Using nitrous oxide (NO) as an indicator, it was possible to calculate the difference between the arterial and venous concentrations to work out how much O2 had
been taken up by the brain. Attempts to apply these techniques to the study of human functional neuroanatomy were largely unsuccessful, as an obvious disadvantage of this method was that only global flow could be measured. Animal studies using autoradiographic techniques were more successful (for review see Sokoloff et al., 1977). However, these techniques rely upon sacrificing the subject at the end of the experiment: thus the procedure does not lend itself to ethical human experiments.
The early work of Kety and Schmidt was subsequently refined by a number of investigators. Ingvar and Lassen were the first to develop methods which took Kety and Schmidt’s earlier work to its logical conclusion - the ability to measure regional CBF in humans (reviewed in Lassen et al., 1991). Building on this work. Glass and Harper utilised Xe-133 (a gamma ray emitter) and used intravenous injections of the tracer and multiple external detectors to develop a much less invasive method. This was then extended with the introduction of computerised tomography to develop Single Photon Computed Tomography (SPECT). The main attraction of SPECT amongst the other imaging techniques available to a 21®^-century neuroscientist is its cheapness: it will not be discussed further.
PET has better temporal and spatial resolution than SPECT because the radionucleotides used in PET are positron emitters, resulting in two gamma rays per disintegration. They have short half-lives, especially those that substitute for stable biologically relevant elements such as carbon, nitrogen or oxygen, and thus allow a number o f images to be collected from each subject. This latter ability makes PET a good technique for conducting activation studies. Although fMRI is
less expensive and has better temporal and spatial resolution, PET allows one to image radiolabelled neurotransmitters and thus obtain a more direct metric of neuronal activity (e.g. Cherry and Phelps, 1996). However, to date there have been few studies that have successfully overcome the complications of modelling the tracer kinetics of neurochemical processes, let alone been able to apply these in a cognitive activation paradigm. The recent work by Koepp and colleagues is a notable exception (Koepp et al, 1998).
2.2.1 H PET: A Brief Overview
As the experimental results presented in this thesis were collected using fMRI, the theory and practice of PET will be discussed in less detail. However, the theoretical issues underlying flow-activation coupling and statistical modelling in PET are maturer than comparable concerns in fMRI, and so serve as a useful introduction.
2.2.1.1 The PET Camera
While there are no theoretical limits on the choice of tracers used for human and animal brain mapping, there are a number of practical considerations to acknowledge. These include the type of emitter used, the half-life of the radioisotope, and the energy of the emitted particles. We will concentrate on the method pioneered by Frackowiak and colleagues at the Hammersmith hospital in London (1980).
is a positron emitter with a half-life of a little over two minutes. When an atom of decays, it emits a positron and a neutrino. While the neutrino rapidly passes through biological material without any interactions, the positron loses energy through encounters with negatively charged electrons. Eventually, the positron will annihilate with one of the electrons. The distance that the particle travels before annihilation depends on the energy o f the positron emitter (Derenzo, 1979). This distance will ultimately affect the spatial resolution of the technique: the further that the positron travels from the tracer, the lower the spatial resolution possible. In human tissue the positron typically travels less than 2mm before annihilation (Cherry and Phelps, 1995).
The signal in PET is not the positron itself, but rather the gamma rays that are produced by the positron/electron reaction. The PET ‘camera’ comprises a ring of detectors surrounding the object being imaged: in the case of human brain mapping, the subject’s head. In modem PET cameras there are a number of detector rings, each one comprising an imaging ‘plane’. The more rings, the greater the coverage of the subject’s head in the inferior/superior plane (commonly labelled the ‘Z ’ axis in imaging experiments). Each ring consists of arrays of positron detectors - dense crystalline materials such as bismuth germanate (BGO) or sodium iodide (Nal). The more crystals that one has in each ring, the greater the possible spatial resolution. Modem PET cameras may have as many as 1024 crystals per ring (reviewed in Roland, 1993).
The materials used in modem PET cameras scintillate when they encounter a gamma ray: in a similar fashion to the processes occurring in scintillation counters, the energy of the gamma rays causes the detector crystals to emit photons. These photons are converted into an electrical signal by a photomultiplier, which uses a photocathode to convert the photons to electrons. The actual signal caused by the interaction of the gamma rays with the crystals is small, however, and so the photomultiplier tubes amplify the signal as well. It is estimated that for every electron caused directly by the interaction of photons at the photocathode, 10^ electrons are produced at the output of the photomultiplier tube (Cherry and Phelps, 1996).
The rings of detectors in the PET camera are designed to register ‘coincidence’ events: as positron annihilation causes two gamma rays to be emitted, ‘tm e’ PET counts are those that cause simultaneous events in two spatially opposed detectors. It should be noted that modem PET cameras allow for coincidence detection between imaging planes, to enable detection of those gamma-ray pairs that are emitted obliquely with respect to the imaging plane. This mode of acquisition is known as 3D data acquisition. It is estimated it offers a five-fold increase in sensitivity over 2D acquisition, in which only coincident events in the same imaging plane are detected (Townsend et al., 1991).
2.2.1.2 Spatial Resolution in PET
In practice, the path between annihilation of the positron and the subsequent detection of gamma rays by the PET camera is not as error-free as described above. There are two hard constraints on PET scanner resolution. The first is the distance that the positron travels before annihilation. The second is caused by the fact that, as the positron and electron are not completely at rest when they annihilate, the emitted gamma rays rarely emit at exactly 180° to each other. This smooths the PET signal, and in doing so lowers the spatial resolution (Derenzo et al., 1982).
In addition to the above constraints, further artefacts can contaminate PET data. The gamma rays can be partially or completely absorbed by tissue: these events give rise to the phenomenae of scatter and attenuation (reviewed in Roland, 1993). Attenuation can be corrected in an efficient manner by using an external source of positrons: from examining the attenuation of this known source, one can efficiently calculate attenuation coefficients (Ranger et al., 1989). Scatter is caused by interactions between gamma rays and other substances (primarily biological material here) before the rays are detected. If one of the gamma rays is scattered and goes on to strike the ‘wrong’ detector, the ‘true’ position of the event will be lost, and a reduction of image contrast will result. Most modem PET cameras employ some means of scatter correction (e.g. Grootoonk et al., 1992). In addition to problems arising from scatter and attenuation, the presence of background radiation can cause the detection o f false positives due to random coincidences at pairs of detectors. Automatic correction algorithms (Hoffman et al., 1981) ably deal with this final noise source.
2.2.1.3 Image Reconstruction In PET
The process of image reconstmction in PET aims to produce a tomographic representation of the concentration of tracer activity. However, the raw data from the PET scanner represents not point activity, but projection activity: the value of the activity detected by each pair of detectors. Raw PET data is usually represented as a sinogram, a 2D matrix whose dimensions represent the angle of the gamma ray projection (^) and the distance the projection line lies from the
centre of the field of view (r) (Cherry and Phelps, 1995). Each element in the sinogram matrix represents a single coincidence line. The difference in computational complexity between 2D and 3D PET can be easily appreciated by considering the massive increase in the number of sinograms which will result when one includes oblique detection planes.
It is usual to employ a backprojection algorithm to reconstruct an image from sinogram data. Backprojection has been likened to ‘...drawing the floor plan o f a house by looking in the windows’ (Croft, 1986). The more windows (imaging planes/projection lines) that one has the better the floor plan (final image). Just as a greater number of windows would facilitate a better sampling of the form o f the house’s floor, the greater the number of detector rings has a similar effect on the PET image. Here, the usefulness of 3D-acquisition mode can be appreciated.
The backprojection method converts the raw data from a Fourier space representation back into image space (Brooks and De Chiro, 1976). In effect, each element of the sinogram is backprojected to the line it represents in image space. The data are usually filtered before backprojection to improve the signal: noise ratio, although this will decrease in the spatial resolution of the data. Backprojection methods for fMRI image reconstruction will be discussed in more detail in the next section.