Elsevier

Molecular Imaging & Biology

Volume 6, Issue 1, January–February 2004, Pages 34-46
Molecular Imaging & Biology

Investigation of a new input function validation approach for dynamic mouse microPET studies

https://doi.org/10.1016/j.mibio.2003.12.002Get rights and content

Abstract

Purpose

Image-derived input functions are desirable for quantifying biological functions in dynamic mouse micro positron emission tomography (PET) studies, but the input function so derived needs to be validated.

Conventional validation using serial blood samples is difficult in mice. We introduced the theoretical basis and used computer simulations to show the capability of a new approach that requires only a small number of blood samples per mouse but uses multiple animals.

Procedures

2-Deoxy-2-[18F]fluoro-D-glucose (FDG) kinetics (60 minutes) were simulated for 10 to 20 animals with three to six blood samples available per animal. Various amounts/types of noise/errors in the blood measurements were assumed, and different amounts/types of errors were added to the true input function to simulate image-derived input function. Deviations between blood samples and the derived input function were examined by statistical techniques to evaluate the capability of the approach for detecting the simulated errors in the derived input function.

Results

For a total of 60 blood samples and a 10% measurement noise, a 5% contaminating error in image-derived input function can be detected with a statistical power of ∼0.9 and with a 95% confidence. The power of the approach is directly related to the error magnitude in the image-derived input function, and is related to the total number of blood samples taken, but is inversely related to the measurement noise of the blood samples.

Conclusion

The new validation approach is expected to be useful for validating input functions derived with image-based methods in dynamic mouse microPET studies.

Introduction

In dynamic micro positron emission tomography (PET) studies,1., 2. a valid time-activity curve (TAC) of the injected tracer in blood, called the input function, is required for reliable quantification of tissue biological function in terms of absolute biological units.3 Image-based methods for deriving input function if validated can make the quantification reliable and practical.4., 5. Even though a multiple of such methods have been validated and used for human studies,4., 5., 6., 7., 8., 9., 10., 11., their applicability to mouse microPET studies that have unique characteristics has not been demonstrated. For example, using factor analysis with positivity constraint,6., 12., 13., 14. we have obtained a blood curve in a dynamic mouse microPET study, as shown in Figure 1. Even though the curve appears reasonable, experimental evidence is needed to characterize the curve’s accuracy/limitation, so one can use the method with full confidence that it will give reliable input functions. Experimental validation of results obtained from such derivation methods for mouse studies is thus critically needed. The conventional validation method, commonly used for studies in humans and large animals, is not applicable for mouse studies, because it uses serial blood samples taken during a dynamic study as a gold standard.4., 6., 7., 8., 9., 10., 11., 15. The procedure of taking serial blood samples is technically difficult due to the small total blood volume in mouse and long time delays in drawing consecutive samples. While, eventually, the difficulty will be solved by some future technical advancement, we have sought alternative approaches that are applicable today.

We have conceived a novel validation approach to address the problem. The approach involves taking only a few blood samples per mouse and measuring their radioactivity concentrations during a dynamic microPET study. After a multiple of separate mouse studies, a large set of deviations between the blood sample measurements and image-derived input function are obtained. The accuracy/adequacy of the derived input functions can then be assessed statistically based on these deviations. In this paper, we present the theoretical basis of the approach, the associated physical, biological, and statistical issues involved, as well as computer simulations that demonstrate the capability of the new approach for discriminating inaccurate input functions. Implication of the approach in terms of the ultimate goal of quantitating tissue biological function will also be discussed.

Section snippets

Theoretical foundation of the new approach

Let Cm(ti) denote a measurement at time ti of a blood TAC (Cb(t)). If the measurement is unbiased, deviation of the measurement from the true blood TAC at the corresponding time point (i.e., Cm(ti)−Cb(ti)) will be a random variate of zero mean value. Let Cd(t) denote the image-derived input function. The deviation of the measurement from the image-derived input function (i.e., ed(ti) = Cm(ti)−Cd(ti)) would have a zero mean value if the derived input function matches the true blood TAC, regardless

Methods

Computer simulation was performed to evaluate the capability of the new approach. For each condition, three to six blood samples (ns) from each of 10 to 20 animals (na) were simulated. In order to have a high statistical confidence of the evaluation results, each experimental condition was repeated 500 times. Details of the computer simulation are described below.

Results

In Figure 2A, a set of 20 simulated input functions in one computer simulation study is shown. The spread among the different curves demonstrated that interanimal and interstudy variability in real studies was considered in the simulation study. Figure 2B shows a derived curve with an inaccuracy simulating a 5% contamination (i.e., r2 = 0.05) from uptake kinetics in tissue. The curve without inaccuracy is also shown as a comparison. The curves show visually the amount of inaccuracy of the derived

Discussions

Using serial blood samples as a gold standard for validating blood curves in a mouse microPET study is hindered by the small blood volume in mouse and the long time delays in drawing consecutive blood samples in mouse studies. At the same time, due to biological and experimental variability (variable injections), it is unrealistic to assume a common input function for different animals or studies, and one cannot simply group together all blood samples from different animals as a single input

Conclusions

The study has shown that the new approach that needs only a few blood samples per mouse can detect small systematic errors in image-derived input functions with high discriminating power. The approach is thus expected to be useful for certifying the accuracy of input functions derived with image-based methods from dynamic mouse microPET studies. The approach can also provide useful information for improvement/refinement of image-based input function methods. The ability to estimate an effective

Acknowledgements

This research was partially supported by the Office of Science (BER), U.S. Department of Energy, Cooperative Agreement No. DE-FC03-02ER63420, and by University of California Biotechnology STAR Grants 01-10184 and 01-10186.

References (24)

  • A. Chatziioannou

    PET scanners dedicated to molecular imaging of small animal models. Mol

    Imaging Biol

    (2002)
  • D. Feng et al.

    Models for computer simulation studies of input functions for tracer kinetic modelling with positron emission tomography. Int

    J. Biomed. Comput

    (1993)
  • S. Cherry et al.

    MicroPET: a high resolution PET scanner for imaging small animals

    IEEE Trans. Nucl. Sci

    (1997)
  • S.C. Huang et al.

    Principles of tracer kinetic modeling in positron emision tomography and autoradiography

  • I.N. Weinberg et al.

    Validation of PET-acquired input functions for cardiac studies

    J. Nucl. Med

    (1988)
  • H. Wu et al.

    Quantification of myocardial blood flow using dynamic N-13 ammonia PET studies and factor analysis of dynamic structures

    J. Nucl. Med

    (1995)
  • H. Wu et al.

    Factor analysis for extraction of blood time activity curves in dynamic cardiac PET studies

    J. Nucl. Med

    (1995)
  • H. Iida et al.

    Use of the left ventricular time-activity curve as a noninvasive input function in dynamic oxygen-15 water positron emission tomography

    J. Nucl. Med

    (1992)
  • M. Graham

    Physiologic smoothing of blood time-activity curves for PET data analysis

    J. Nucl. Med

    (1997)
  • C.J. Hoekstra et al.

    On the use of image-derived input functions in oncological fluorine-18 fluorodeoxyglucose positron emission tomography studies. Eur

    J. Nucl. Med

    (1999)
  • J.S. Lee et al.

    Robust extraction of input function from O-15 water dynamic myocardial positron emission tomography using independent component analysis

    IEEE Trans. Nucl. Sci

    (2000)
  • H. Watabe et al.

    Noninvasive estimation of the aorta input function for measurement of tumor blood flow

    IEEE Trans. Med. Imaging

    (2001)
  • Cited by (27)

    • Estimation of the net influx rate K<inf>i</inf> and the cerebral metabolic rate of glucose MR<inf>glc</inf> using a single static [<sup>18</sup>F]FDG PET scan in rats

      2021, NeuroImage
      Citation Excerpt :

      Although this procedure is feasible in rats (Hori et al., 2015; Moore et al., 2000; Shimoji et al., 2004), measuring the arterial input function without affecting the animal's physiology remains challenging. For this reason, alternative approaches using population-based (Meyer et al., 2006; Takikawa et al., 1993) or image-derived (Alf et al., 2013; Hori et al., 2015; Huang et al., 2004; Kudomi et al., 2011) input functions have been developed. Nonetheless, they still require extensive dynamic acquisitions (60–90 min), thereby considerably reducing scan throughput and they are not compatible with an awake uptake period.

    • Kinetic modeling in pre-clinical positron emission tomography

      2014, Zeitschrift fur Medizinische Physik
      Citation Excerpt :

      Then only few or small samples (some μl) can be extracted, which has the potential to introduce several errors such as error in pre- and post-weighing of the small samples and also the higher noise contribution due to low counts in the gamma-counter (especially for C-11 studies and late time points). Apart from manual sampling, various dedicated automated blood-sampling devices have been introduced in recent years [44–47] which offer the possibility of measuring small samples with accurate volume and good time determination and which can be used for rats and mice. Another invasive method is the use of beta-probes [48–51] to measure the input function.

    • Towards a reproducible protocol for repetitive and semi-quantitative rat brain imaging with <sup>18</sup> F-FDG: Exemplified in a memantine pharmacological challenge

      2014, NeuroImage
      Citation Excerpt :

      However, pharmacokinetic modelling requires invasive catheterization procedures to obtain a plasma blood input function (Alf et al., 2013; Meijer et al., 2005). Therefore alternatives employ an image-derived input function by a dynamic scan of the heart (Huang et al., 2004; Kemp et al., 2009; Kim et al., 2006; Kudomi et al., 2011; Laforest et al., 2005; Locke et al., 2011) or liver (Hudson and Larkin, 1994; Kudomi et al., 2009) or make use of population based approaches (Defrise et al., 1997; Takikawa et al., 1993). Nevertheless, to obtain an accurate measure from the kinetic modelling, scan durations of 60 up to 90 min need to be considered (Huang et al., 1980; Watson, 2000), greatly reducing scan throughput and limiting feasibility.

    • Molecular imaging in neuroscience research with small-animal PET in rodents

      2011, Neuroscience Research
      Citation Excerpt :

      Friston et al. (2007) developed SPM, a data analysis tool that is widely used in brain function analysis to identify specific regional effects in neuroimaging data and to characterize functional anatomy, specialization, and disease-related changes in the human brain. Many methods have been also established for quantifying regional changes in the rodent brain with small-animal PET (Rubins et al., 2003; Huang et al., 2004; Laforest et al., 2005; Kim et al., 2006; Schiffer et al., 2007). The extension of these techniques to an automated whole-brain imaging analysis method, either by predefined volume-of-interest (VOI) analysis or by SPM analysis, has been systematically investigated in small animals (Schweinhardt et al., 2003; Casteels et al., 2006; Mirrione et al., 2006, 2007; Hjornevik et al., 2007).

    • Small-Animal Preclinical Nuclear Medicine Instrumentation and Methodology

      2008, Seminars in Nuclear Medicine
      Citation Excerpt :

      Using this method, the authors found quantification difficult but were able to measure an accurate RIF with high temporal resolution. Another method sought to obtain the RIF using minimal blood sampling combined with computer simulation to validate image derived RIFs.78 RIFs for 10 to 20 animals injected with FDG were simulated while also taking 6 blood samples per animal.

    • Long-Term Monitoring of Transplanted Islets Using Positron Emission Tomography

      2006, Molecular Therapy
      Citation Excerpt :

      These advances increase the image quality through reduction in spillover effects and make the imaging systems more capable of detecting lower signals. Advances with PET–CT combinations for attenuation correction, together with recent work to generate probe concentrations in blood from image data [34,35], will make possible quantitative estimates of metabolic rate constants that improve the ability to see early changes in the islet metabolism. The combination of new probes and imaging technologies may provide powerful new tools to monitor and prolong the survival of islets in individuals at risk for, or with, type 1 or 2 diabetes.

    View all citing articles on Scopus
    View full text