Positron emission tomography (PET) is now a well-established molecular imaging methodology, able to visualise and quantify function in the brain, heart and whole body. However, the full potential of PET lies in the diverse array of radiotracers that can be used, whereby specially designed radiotracers can be used to examine a wide and diverse array of processes in vivo, including for example, glucose metabolism, perfusion, neurotransmission, hypoxia, gene expression and cell tracking. PET has been largely conducted with only one radiotracer being imaged at a time, meaning that images of different processes are acquired separately (if more than one scan is even conducted). Examples of multiple clinical scans, conducted separately, include [18F]FDG PET to look at glucose metabolism, and a subsequent separate [11C]methionine scan (for imaging of protein synthesis, to find the boundaries of a brain tumour, for example).
There is clearly scope for more powerful characterisation of disease and diagnoses by imaging multiple tracers simultaneously, or at the very least within the same single PET imaging session. Such an approach would open up new clinical imaging and research possibilities, for example by imaging perfusion and hypoxia simultaneously in the heart (and hence separating these measures), or, as another example, by interrogating multiple neurotransmission systems at once in the brain.
However, multi-radiotracer administration has largely been avoided, due to the obvious problematic overlap in the radiotracer distributions, particularly when the physical or biological half lives are comparable. However, with deep learning techniques, and the ability to train artificial neural networks (ANNs) with extensive examples in order to disentangle complex mixtures of data, this project ambitiously seeks to explore the frontiers and full potential of multi-radiotracer imaging for PET.
Even if ambitious, the feasibility of this project is nonetheless assured, as it has already been demonstrated that reconstruction and analysis of dual tracer imaging is possible (e.g. ), using simpler independent component analysis (ICA) methods. It is therefore anticipated that by exploiting machine learning there will be scope for improving upon previous work and imaging more tracers, even more so if machine learning is also combined with carefully designing and optimising the input functions for each radiotracer (characterised by total injected dose (area under the curve), injection rate and injection time). By optimising these three key parameters for each radiotracer, and furthermore through use of deep learning neural network architectures, it is expected that more than 2 radiotracers will be able to be imaged simultaneously (where, for example, the injection time parameters would cover the case of rapid sequential imaging in a single scan). Recent advances in low-dose image reconstruction, using for example the kernel method, demonstrate the feasibility of multi-tracer imaging also from a radiation dose perspective. Therefore the full potential remains wide open for research. Hence the overall aim of the work will be a systematic investigation into the extent of the potential for multi-tracer imaging, by optimisation of the administration strategy and also the signal separation method through use of extensive simulations.
It is expected that this work will not only open up the potential for multiplexed PET imaging using existing PET scanners and tracers, but furthermore that it will form essential prerequisite work for realising the full potential of new radiotracers and in particular total body PET scanners, which may find application in screening for disease, as well as improved diagnoses, and ability to monitor therapies.
 J. Verhaeghe, and A. J. Reader, “Simultaneous water activation and glucose metabolic rate imaging with PET,” Phys Med Biol, vol. 58, no. 3, pp. 393-411, Feb 7, 2013.