In parallel, 18F-FDG imaging of brain tumours provides information regarding the localization and malignancy of the lesion. The ability to perform [15O]H2O and FDG PET simultaneously would enable the study of water activation and glucose metabolic rate in one scan, while avoiding the need for image co-registration and reducing motion problems.

With this aim, Jeroen Verhaeghe and Andrew Reader from the Brain Imaging Centre at McGill University, Canada, have proposed a strategy for separating FDG and multiple [15O]H2O signals from a simultaneously acquired dynamic PET acquisition (Phys. Med. Biol. 58 393).

Separation strategy

FDG and [15O]H2O exhibit highly distinct time activity curves (TACs), enabling the use of temporal information to separate the tracer signals. To do this, the researchers defined two sets of temporal basis functions, which model either the slowly decaying FDG or the fast decaying water dynamics. These basis functions are formed by convolution of a generating function, estimated from the measured head curve, with a set of pre-determined decaying exponential functions.

"The main difference of our approach, compared to previous dual-tracer techniques, is the exploitation of a temporal basis function methodology, as used in fully 4D image reconstruction," Verhaeghe explained. "Such an approach effectively permits every time point in the reconstructed image to benefit from far more of the acquired data than a conventional frame-by-frame approach."

Verhaeghe and Reader performed extensive PET simulations of a 3D brain phantom during three repetitions of baseline and task conditions. During each task, a region located in the lingual gyrus was simulated to have increased blood flow. The total acquisition time was 75 minutes, with a single FDG injection at the start, followed by [15O]H2O injections at 10, 21, 32, 43, 54 and 65 min.

Activity distributions for FDG and [15O]H2O were simulated independently, and the two sinograms added and reconstructed. The researchers then used their proposed separation method to isolate the FDG and [15O]H2O components from the reconstructed images. For comparison, they also studied a simpler separation technique based on linear interpolation of the FDG component at times when [15O]H2O activity was negligible.

Improved performance

Reconstructing the FDG and [15O]H2O TACs using the basis functions method resulted in reduced noise compared with the use of linear interpolation, though not necessarily also a reduced bias. "The overall error, however, which reflects error due to noise and bias, is always lower when using the basis function approach," Verhaeghe added.

The researchers created parametric maps for the two separation techniques, as well as for individually measured tracers. For FDG, linear interpolation increased the noise compared with individual imaging, while the basis function method reduced the noise compared with linear interpolation.

Maps of [15O]H2O activation created using the basis functions method showed increased t-values in the activated region (indicating that the difference between activated and baseline conditions is genuine rather than noise-related). With linear interpolation, the elevated values were harder to distinguish from the noise.

Verhaeghe and Reader also assessed errors in calculated activity values for four regions-of-interest (ROIs) – located in the frontal cortex, caudate, white matter and the activated region – using 50 pixels per ROI.

For FDG, simultaneous imaging with linear interpolation increased the error in all ROIs compared to individual imaging. In contrast, simultaneous imaging using the basis function method reduced noise to below that of conventional individual imaging, which is a notable achievement. The basis function method underestimated true FDG activity by 1.5% and 2.4% in the frontal cortex and the activated region, respectively, and overestimated the value in white matter by 1.1%. Simultaneous imaging using linear interpolation always exhibited the largest mean absolute error (MAE), for example, 8.4% in the white matter.

For the static [15O]H2O component, the basis function method overestimated activity in all cases, and MAE increased for simultaneous (rather than individual) imaging. For non-activated regions, all methods produced t-values close to zero. In the activated region, a slightly lower t-value was seen when using the basis function method versus individual imaging, with a further reduction when using linear interpolation.

Verhaeghe and Reader concluded that dual-tracer imaging using basis functions to separate FDG and [15O]H2O components is feasible, and that the resulting images and values are comparable to those obtained from individual PET scans. The linear interpolation method was deemed less suitable as it significantly increased the noise. They now plan to test this method on actual acquired PET data, where the method could prove of considerable benefit in preparing for brain surgery.

Related articles in PMB
Simultaneous water activation and glucose metabolic rate imaging with PET
Jeroen Verhaeghe and Andrew J Reader Phys. Med. Biol. 58 393
Dual-isotope PET using positron-gamma emitters
A Andreyev and A Celler Phys. Med. Biol. 56 4539
Task-oriented quantitative image reconstruction in emission tomography for single- and multi-subject studies
Jeroen Verhaeghe et al Phys. Med. Biol. 55 7263