TY - JOUR T1 - Association of Image-derived Blood Input Function with Patient Characteristics in Dynamic FDG-PET: A Pilot Study JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1315 LP - 1315 VL - 58 IS - supplement 1 AU - Benjamin Spencer AU - Amir Sahabi AU - Andrea Ferrero AU - Michael Rusnak AU - Heather Hunt AU - Andrew Lee AU - Souvik Sarkar AU - Edward Kim AU - Jinyi Qi AU - Ramsey Badawi AU - Guobao Wang Y1 - 2017/05/01 UR - http://jnm.snmjournals.org/content/58/supplement_1/1315.abstract N2 - 1315Objectives: Image-derived input function (IDIF) is often the choice for pursuing tracer kinetic modeling to avoid frequent invasive arterial blood sampling. However, IDIF is difficult to obtain in breast imaging on dedicated high-resolution breast PET scanners because neither the heart nor major blood vessels are in the field of view. We aim to develop a method of predicting patient-adaptive IDIF of dynamic FDG-PET from data of patient characteristics and clinical variables using machine learning. As the first step towards this goal, in this work, we investigate the correlation between the key parameters (e.g. peak value) of IDIF and clinically measured patient variables.Methods: Ten patients undergoing dynamic FDG-PET/CT scan at UC Davis Medical Center were included in this study. Five of these patients were scanned for liver disease, 2 for pancreatic cancer and 3 for breast cancer. All dynamic scans were performed on a GE whole-body PET/CT 690 scanner and lasted for 60 minutes. The scan duration of time frames in the first 5 minutes was 10 seconds. The IDIF of each patient was derived by manual placement of volume of interest (VOI) in the descending aorta regions in the image sequence of standardized uptake values (SUV) using the AMIDE software. The peak of IDIF was then determined as the maximum value in the early phase. We collected nine relevant patient variables including height, weight, gender, age, body mass index (BMI), lean body mass (LBM), blood sugar level, injected dose, and the IDIF value at one hour post injection (bSUV60). The association between the IDIF peak and predictors was analyzed using the univariate Pearson’s correlation coefficient and multivariate linear regression. Leave-one-out cross validation (LOOCV) was further used with regression to evaluate the ability of variables for predicting the IDIF peaks. In addition to the 10-s temporal sampling used in early dynamic scans, we also applied a higher temporal sampling of 5-second in 7 patient data sets (for which list-mode data were available) to better recover the peak value of the IDIF and investigate the effect of temporal sampling on the correlation analysis.Results: The univariate correlation analysis of ten patients showed that IDIF peaks correlated with bSUV60 (r=0.862, p=0.001). The peaks tended to correlate with patient weight (r=0.622, p=0.055) and BMI (r=0.628, p=0.052). No correlations were found between the peaks and blood sugar level (r=-0.071, p=0.846) as well as other variables. In the multivariate linear regression analysis, no variables in addition to bSUV60 were found to be correlated with the IDIF peaks (p>0.1). In the LOOCV analysis, the adjusted R2 of bSUV60 for predicting IDIF peak was 0.575. Addition of one or more other variables decreased the adjusted R2. In the secondary analysis of seven patients, when temporal sampling rate were changed from 10-s to 5-s, the IDIF peaks had a significant increase from 46.22±11.62 to 64.29±7.92 (p<0.001). The correlation of the IDIF peaks with bSUV60 was improved from r=0.856 (p=0.014) to r=0.969 (p<0.001). The correlation improvement tended to be significant (1-tail p value = 0.033 and 2-tail p value = 0.067). The adjusted R2 in the LOOCV for predicting IDIF peaks was also improved from 0.524 to 0.863.Conclusion: A significant correlation existed between the peak of the IDIF and bSUV60 in dynamic FDG-PET/CT studies. The correlation can be improved if 5-s temporal sampling rather than 10-second sampling was used. No strong association were found between the IDIF peak and the patient variables height, weight, gender, age, BMI, LBM, blood sugar level, and injected dose. The result also showed that bSUV60 alone can be a strong predictor for estimating the peak of IDIF. In practice, bSUV60 can be either obtained by a static whole-body PET scan or by a blood draw at the end of the dedicated dynamic breast PET scan. Research Support: This research was supported in part by the CBCRP grant #21IB-0133 and ACS IRG-95-125-13. ER -