TY - JOUR T1 - Towards non-invasive estimate of blood-plasma time activity curve for dynamic PET JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 142 LP - 142 VL - 60 IS - supplement 1 AU - Gregory Romanchek AU - Brett Yockey AU - Shiva Abbaszadeh Y1 - 2019/05/01 UR - http://jnm.snmjournals.org/content/60/supplement_1/142.abstract N2 - 142Objectives: Dynamic FDG-PET imaging has been shown to provide quantitative metrics and more accurate tumor-to-background Contrast-to-Noise Ratio (CNR) over traditional FDG-PET in both single- and multi-bed studies [1-4]. Necessary to dynamic PET, however, is the blood-plasma time activity curve (the input function), from which tissue uptake is derived in accordance with the Patlak compartment model [5]. Acquiring the input function, however, requires periodic probing of activity in blood-pool regions. Traditionally, and in single-bed studies excluding the neck or heart region, timed blood sampling of the patient remains the only option for this measurement [6]. As this is invasive and uncomfortable for the patient, it is not typically used despite its functionality. In multi-bed PET studies, sequential frames over the heart region provide a way to derive the input function through imaging [1]; but this requires multiple frames over the heart region, excluding this protocol from single bed studies or from whole-body single pass studies. The goal of this project is to estimate the relationship between phycological characteristic of the patient such as GFR (Glomerular Filtration Rate) [7], image-derived information in the absence of the heart region, and blood-plasma activity to estimate the input function non-invasively. Methods: In an ongoing clinical study, 11 patients diagnosed with cancers in the pelvic region have undergone traditional whole-body FDG-PET/CT imaging so far. Urine samples were collected prior to imaging for activity measurements (0.534 ± 0.288 mCi). The activity concentrations (AC) of the left-atrium (0.158 ± 0.073 μ Ci/mL), generalized muscle (2.77 ± 1.18 kBq/mL), and bladder (1.965 ± 0.792 μCi /mL) were acquired via PMOD, an oncological physical modeling software [8]. Patient GFR (64.128 ± 18.318 ml/min/1.73m2) was also computed from most recent creatinine testing. Each activity-based measurement was normalized to the injected dose of the patient. From dose-normalized (DN) muscular uptake, DN total urine activity (urine sample plus estimated amount present in bladder), GFR, injected dose, and weight, a second-degree polynomial was constructed. The terms of this polynomial consist of the sum of each metric (5 terms) plus a pairwise multiplication between each metric (25 terms) plus a constant scalar (1 term). The weights were trained via ordinary least-square regression with a regularization factor of 1 against the true DN left atrium AC. For validation, since limited training examples are currently available, the regression was repeated 11 times with each of these trials reserving one of the 11 patient’s data as a test set for computing test Root-Mean-Square Error (RMSE). Results: The test RMSE of the 11 individual trials was 0.0142 ± 0.0139 and the trials are summarized in Table 1. The full sample training and testing agrees with a RMSE of 0.0116 and is plotted below in Figure 1. Conclusions: Blood-plasma activity concentration can be estimated from metrics obtained non-invasively from patient information, image-derived sources from the pelvic region, and GFR. While GFR requires blood sampling, it is expected that any cancer patient will have had a recent test given their health condition. Further testing is required to determine whether the blood-plasma activity can be estimated dynamically from muscular and bladder kinematic uptake. ER -