Statistical tests for discriminating inaccurate input functions in dynamic mouse microPET studies

Image-derived input functions (EDIF) are desirable for quantifying biological functions in mouse microPET studies. Due to difficulties in taking many blood samples from each mouse, conventional IDIF validation method of comparing blood samples with IDIF in a single animal is not applicable. A new approach that requires statistical testing on data of multiple animals has been conceived and investigated for IDIF validation. In this study, we evaluate and compare the power of 5 common statistical tests-Chi-square, sign test, signed-ranks, runs test, serial correlation-with the new approach for their ability to detect errors in IDIF. Computer simulation was used to simulate mouse FDG kinetics (60 min) and error-containing IDIF of various conditions. Deviations of IDIF from blood samples were examined with the 5 statistical tests. Results show that sign test, runs test and serial correlation do not have comparable power as the other two tests. The signed-ranks test generally had high statistical power, but was unable to detect errors that are random among animals or studies. Chi-square test could detect error in IDIF that was variable from study to study, but required accurate knowledge of noise variance. Combining signed-ranks test with Chi-square test is overall most sensitive for validation of IDIF.