Comparison of Image-Derived and Arterial Input Functions for Estimating the Rate of Glucose Metabolism in Therapy-Monitoring ¹⁸F-FDG PET Studies

MATERIALS AND METHODS

RESULTS

MRGlu based on the ascending aorta IDIF (Fig. 1A), the left ventricle IDIF (Fig. 1B), and the abdominal aorta IDIF (Fig. 1C) showed a strong relationship with MRGlu based on arterial sampling. Comparability between the measures was also high, with ICC being 0.98 (95% confidence interval, 0.97–0.99) for the ascending aorta IDIF, 0.94 (0.92–0.96) for the left ventricle IDIF, and 0.96 (0.93–0.98) for the abdominal aorta IDIF. The slopes of the 3 scatterplots showed a slight systematic deviation of 6%–8%; intercepts were near zero (Fig. 1). No differences in distribution were observed between the 3 types of cancer.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Scatterplots of MRGlu, expressed in μmol·mL⁻¹·min⁻¹, as calculated using ascending aorta IDIF (A), left ventricle IDIF (B), or abdominal aorta IDIF (C) to that calculated using arterial-sampling–derived input function. Ninety-five percent prediction intervals are indicated by dotted lines.

Comparability for MRGlu based on the left ventricle IDIF and the ascending aorta IDIF was 0.95 (95% confidence interval, 0.92–0.97). For MRGlu based on the left ventricle IDIF and abdominal aorta IDIF, the ICC was 0.95 (0.90–0.97).

Figure 2 shows the scatterplots of the MRGlu based on the IDIF of the left ventricle in patients without (Fig. 2A) and with (Fig. 2B) myocardial ¹⁸F-FDG uptake, versus arterial-sampling–based MRGlu. For the group without myocardial ¹⁸F-FDG uptake (n = 59), the ICC was 0.95 (0.92–0.97). The ICC for the group with myocardial ¹⁸F-FDG uptake (n = 39) was 0.94 (0.89–0.97).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Scatterplots of MRGlu values based on IDIF of left ventricle vs. MRGlu values based on arterial sampling, for groups without (A) and with (B) ¹⁸F-FDG uptake in myocardial wall.

DISCUSSION

MRGlu based on IDIFs from the ascending aorta, the left ventricle, and the abdominal aorta showed highly significant correlations with MRGlu derived from arterial blood sampling, which served as the reference method. The ascending aorta IDIF did not show any prominent outliers and had the strongest correlation with the gold standard. The fact that the ICCs for MRGlu estimated by all 3 types of IDIFs compared well with those for arterial-sampling–derived MRGlu suggests that the use of uncorrected IDIF methods provides consistent data that can be translated to clinical practice. The advantage of using the ascending aorta is the possibility of defining larger VOIs, thus obtaining data with better statistical properties and less spillover from potentially high-uptake tissues such as myocardium. Partial-volume effects, however, remain possible.

In studies monitoring therapeutic response, IDIFs from large vessels or the left ventricle are already used (9,10). However, the data are rarely validated against the gold standard. Although the validity of using IDIFs has previously been investigated in several studies of PET radiopharmaceuticals other than ¹⁸F-FDG (2–4,11,12), similar studies using ¹⁸F-FDG have never been performed on large numbers of oncologic patients.

Our findings are in line with the findings of van der Weerdt et al. (1), who evaluated the accuracy of IDIF from the left ventricle, left atrium, and ascending aorta in 18 cardiac ¹⁸F-FDG PET studies. However, the left atrium is difficult to localize in oncology studies because of poor contrast after 6 h of fasting, and potential spillover from the myocardium is less than in cardiac ¹⁸F-FDG PET. Therefore, results obtained from such cardiac studies (1,6,13) cannot be directly extrapolated to ¹⁸F-FDG PET for oncologic applications.

A number of factors can affect the accuracy of IDIFs. IDIFs potentially suffer from partial-volume effects. Several methods have been reported for correcting time–activity curves for partial-volume effects—for example, estimation of the diameter of the vessel via activity profile analysis with nonlinear regression techniques (14). Furthermore, the left ventricular curve can contain spillover from the myocardium, because of the limited spatial resolution of PET scanners and cardiac movement. This spillover can affect the tail of the input curve and lead to errors in glucose consumption. The errors could be corrected by using information from independent venous blood samples in later time frames (15). IDIFs are also susceptible to errors due to patient movement. In addition, noise could be introduced by the limited number of counts acquired in each time frame. Especially small VOIs suffer from higher noise levels. The dependence of the final recovery coefficient on the filter used in the image reconstruction process can be determined analytically (14). The blood volume in tumor tissue modulates uptake of the tracer. Therefore, the use of blood volume corrections can influence diagnostic accuracy (16). Fractal dimension may help to quantify heterogeneity. An increased fractal dimension indicates a more chaotic distribution of ¹⁸F-FDG (16). The application of factor analysis, a method of generating an input function from smaller vessels, also appears promising but needs further evaluation (17). Furthermore, manual VOI definition of vessels could introduce inter- and intraobserver variability.

In general, arterially sampled and uncorrected IDIFs will differ. When only the partial-volume effect is important, the IDIF curve will be isomorphic to the arterially sampled curve (Fig. 3A). For all time points, the IDIF activity concentrations will be lower than the arterially sampled ones by the same factor f (f < 1), resulting in an IDIF-derived MRGlu that is higher by 1/f than the MRGlu derived from arterial sampling. Conversely, when trapping of ¹⁸F-FDG occurs, such as in the myocardium or in blood vessel walls, and partial-volume effects are small, the IDIF and arterially sampled values will be equal up to the time at which the relative contribution of trapped ¹⁸F-FDG can be neglected. At later times, however, the IDIF values will exceed the arterially sampled ones (Fig. 3B). This trapping effect will lead to lower MRGlu values for IDIFs than for arterial sampling. Obviously, in clinical practice, both effects may be present, leading to slope values greater than 1 when the partial-volume effect is predominant and to slope values less than 1 when trapping is predominant. It is clear that for both aorta-derived input functions, the partial-volume effect dominates (slopes > 1), whereas for the left ventricle-derived input functions, the trapping effect is stronger. This effect is further exemplified by the fact that MRGlu values based on left ventricle–derived input functions lead to a slope of 0.98 when myocardial ¹⁸F-FDG uptake is absent and to a slope of 0.88 when myocardial uptake is high.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

IDIF left ventricle of a patient without myocardial uptake of ¹⁸F-FDG (A) and a patient with high myocardial uptake of ¹⁸F-FDG (B) plotted against arterially sampled input functions derived from same dynamic dataset. In A, only partial-volume effect affects IDIF. All time points are lower, but IDIF curve stays isomorphic to arterial-sampling curve. In B, trapping of ¹⁸F-FDG is predominant. Curves stay isomorphic until time at which trapping cannot be neglected and IDIF values start to exceed arterial-sampling values.

CONCLUSION

Despite a variety in potential errors when generating IDIFs, strong correlations were observed between MRGlu based on IDIFs and MRGlu based on arterial sampling. The use of uncorrected image-derived methods is accurate and simple, can be applied to studies monitoring oncologic therapy, and represents a clinically viable alternative to arterial blood sampling.