Arterial input function from PET image
Absolute quantitative PET studies require that the input function (IF), representing cumulative availability of the radiotracer in arterial plasma (AIF), is measured. Traditionally this is achieved via arterial cannulation. Image-derived input function (IDIF) is a noninvasive alternative to arterial blood sampling, but also associated with several problems that must be solved and methods well validated before using in clinical studies (Asselin et al., 2004; Zanotti-Fregonara et al., 2011; Christensen et al., 2014). IDIF can only be implemented with a minority of PET tracers, and even then some blood samples usually need to be taken (Zanotti-Fregonara et al., 2011), unless some other form of calibration is possible (Christensen et al., 2014; Zanotti-Fregonara et al., 2014). Injection catheter can be used for venous blood sampling at late times (Hoekstra et al., 2000). Numerous research publications contain validation of a specific IDIF method for a specific study protocol, but the methods rarely have been applicable in other study protocols or research institutes without laborious revalidation.
Arterial blood curve (BTAC) is relatively easy to measure from heart cavities and ascending and descenting aorta because of their large size in humans (Henze et al., 1983; Weinberg et al., 1988; Gambhir et al., 1989; Ohtake et al., 1991; van der Weerdt et al, 2001; de Geus-Oei et al., 2006; Horsager et al., 2015), if those are visible in PET image. Quality control of LV cavity derived BTACs should be performed to prevent large errors (Hoekstra et al., 1999). Probably the most successful application is the method for quantification of myocardial perfusion using [15O]H2O (Iida et al., 1991 and 1992). When heart is not in the field of imaging, other large arterial blood pools that can be used to estimate arterial blood curve are abdominal aorta (Ohtake et al., 1991; Germano et al, 1992), femoral arteries (Lüdemann et al, 2006; Croteau et al., 2010), and common iliac arteries (Schiepers et al., 2008; Cho et al., 2010). Image reconstruction method may affect the quality of the IDIF; OS-EM can reduce noise, but cause bias, when compared to FBP (Feng et al., 2015).
If a region which could be used to extract decent AIF does not fit into the FOV of the PET scanner, a common practice in clinical PET studies is to first position the subject so that heart is in the FOV, and after the arterial peak has passed (usually after 10-20 min), bed is moved and the late scan of the region of interest is started; the initial BTAC is measured from small ROI placed in LV cavity of the heart, and it is combined with venous samples withdrawn during the late scan (Eary & Mankoff, 1998). With very short-lived radionuclides and standardized injection protocols, a repeated PET study can be performed, scanning first the heart for the input, and then the region of interest, using the input function from the first scan either directly or after scaling (Tolbod et al., 2018). In a more advanced (and expensive) setting, another PET scanner could be used to scan for example heart at the same time as the tissue uptake is measured for example in the brain (Iida et al., 1998). Also wrist scanners have been studied with the aim of noninvasive measurement of arterial blood curve from the radial artery (Aykac et al., 2001; Shokouhi et al., 2003; Kriplani et al., 2007a, 2007b).
Suboptimal reconstruction methods may lead to substantial bias in estimated IDIF (Boellaard et al., 2001). Arterial inflammation, for example in chronic kidney disease, may lead to IDIF overestimation, if the tracer is avidly taken up by inflammatory cells (for example [18F]FDG) or arterial calcifications (for example [18F]F-).
IDIF methods at their best can provide the arterial blood curve, which for most tracers need to be further converted to plasma curve and corrected for circulating label-carrying metabolites. (Nahmias et al., 2000; Mourik et al., 2009).
In brain PET studies the blood curve is estimated from carotid arteries or other intra-cranial blood vessels, but because of large partial volume effect and proximity of large veins the results may be poor and require a few blood samples for calibration. Different approaches like factor analysis, non-negative matrix factorization (Kim et al., 2001), independent component analysis (ICA), cluster analysis, and applying MRI (Fung et al., 2010; Fung & Carson, 2013; Iguchi et al., 2013; Su et al., 2013; Lyoo et al., 2014; Mabrouk et al., 2014; Simončič & Zanotti-Fregonara, 2015; Jochimsen et al., 2016; Khalighi et al., 2018; Kang et al., 2018) have been used to overcome these problems. Patient motion must be minimal or corrected for (Mourik et al., 2011). Hahn et al 2012 have proposed using linear discriminant analysis (LDA) to delineate vessels from the image. Integrated PET-MRI allows MR angiography for segmenting the carotid arteries and monitoring subject movement (Islam et al., 2017; Su et al., 2017; Sundar et al., 2018).
Mean diameters of internal carotid artery and common carotid artery are 5.11±0.87 mm and 6.52±0.98 mm in men, and 4.66±0.78 mm and 6.10±0.80 mm in women (Krejza et al., 2006). Croteau et al (2010) measured carotid diameter from CT image, and used phantom based recovery coefficient and spill-in factors to correct carotid blood curves. Similar approach was used by Feng et al (2012).
Schain et al. (2013) identified the voxels belonging to the carotid by computation of the Pearson product-moment correlation coefficient between a seeding region and voxels in a larger region (excluding voxels in the seeding region). First, coordinates of the highest local maximum are automatically extracted from time frame where the carotids are most clearly visible. Coordinates are provided to the MATITK Connected Threshold Segmentation method to obtain a seeding region. Seeding region is dilated to obtain the subset mask. The final voxels contributing to the IDIF are extracted using the Pearson correlation method. The extracted curve is then scaled to blood radioactivity using blood samples (Bahri MA et al., 2017).
IDIF methods in small-animal studies are especially important, because reliable and frequent blood sampling may not be possible (Laforest et al., 2005). Heart cavity and vena cava (Yee et al., 2005; Lanz et al., 2014) can be used, if appropriately validated. Vena cava is problematic, if tracer was injected into tail vein. In FDG mice studies, about 2 min p.i., vena cava provides more reliable AIF than heart cavity (Thorn et al., 2013; Thackeray et al., 2015). Blood samples taken from the tail or femoral artery could be combined with initial data from the LV cavity (Shoghi & Welch, 2007; Wong et al., 2011). Initial BTAC from LV cavity could also be combined with scaled TAC of the liver (Tantawy & Peterson, 2010). Clustering methods can be helpful in automating IDIF estimation from mice PET images (Zheng et al., 2011). Integrated PET-MRI helps to perform partial volume correction for mouse heart (Evans et al., 2015). Factor analysis of rat or mouse heart can separate the TACs of myocardium and blood (Wu et al., 1996; Laforest et al., 2005; Herrero et al., 2006; Kim et al., 2006). Modelling of the combined kinetic information from myocardial muscle and LV cavity is another option to extract the input function (Fang & Muzic Jr, 2008; Kudomi et al., 2011; Xiong et al, 2012; Li et al., 2015). Gated PET imaging (Locke et al., 2011; Mabrouk et al., 2012; Zhong & Kundu et al., 2013a and 2013b; Evans et al., 2015) can improve the separation of myocardial muscle from the cavity.
Perfusion in the liver is high, and blood-to-hepatocyte transfer is practically unlimited for most radioligands; therefore radioactivity concentration in the liver follows relatively well the BTAC. [18F]FDG concentration in the liver has been suggested to be an suitable substitute for BTAC in small animal studies (Green et al., 1998; Yu et al., 2009; Tantawy & Peterson, 2010) and even in diagnostic PET (Torizuka et al., 1995).
After bolus administration of the radioligand, the radioactivity concentration in the arterial blood changes rapidly. To accurately follow the kinetics of the input_function, blood sampling should be frequent in the initial phase of the study. Compartmental model fitting is reliable only when the kinetics of the input function and tissue data are accurately measured. However, the in vivo measurement of radioactivity requires relatively long time frames to achieve sufficient count statistics. The longer the time frames, the less kinetic information can be collected, but the measured concentrations are less noisy.
As the peak of the input curve is usually very sharp (on purpose), it seems to be flattened when measured as an average during PET frames. The effect is pronounced when the peak time is divided between two adjacent time frames, in comparison to when the peak resides mostly during a single time frame:
Detecting the position of the blood vessel in the image is usually (depending on the tracer) only possible right after the administration, at the time of the blood peak. Towards the end of the study, the activity concentration in the surrounding tissue reaches and often exceeds the activity in the vessel. Therefore the detection of the blood vessel may not be possible from late PET scans. Even in dynamic scan, if the initial time frames are long, the blood peak may not be visible in the image data. Additionally, the obtained blood TAC may look like the peak was missed (Figure 2), while the AUCs of the TAC may still be correct and usable in Patlak plot.
Locating the artery in the dynamic image
Summing of only the first few frames of the dynamic image reveals the position of large arteries. Locating the arterial VOI can be automated by calculating for each pixel the difference between mean activity during first few minutes and during the end frames, and then finding a fixed-sized VOI where this value is at its maximum (Backes et al., 2009).
Seeded region growing and cluster analysis have also been used to identify arterial VOIs (Liptrot et al., 2004; Mourik et al., 2008). Pairwise correlations between image voxels can be used to identify pixels with similar kinetics (Schain et al., 2013).
Correcting vessel activity for PVE
VOI drawn on a small vessel contains activity of blood and surrounding tissue (background) because of the partial volume effect (PVE). If VOI is small, then part of the blood activity is spread outside of the VOI, but if VOI is large, then we can assume that no blood activity is spread outside of it. In that case, the activity concentration in the VOI is the weighted sum of arterial and background concentrations (CA and Cbg, respectively)
, where the weighting factor fv is the fractional volume of the vessel inside the VOI. Backes et al (2009) assumed for [18F]FLT that the ratio of background-to-arterial concentration can be described by exponential function with rate constant k:
, and thus the true arterial concentration could be calculated from equation (Backes et al., 2009):
Additionally, they noted that arterial blood TAC could be well fitted with a seven-parameter function. The model that consists of the two compartmental model parameters and seven function parameters was then fitted to the TAC of the carotid ROI. Objective function contained also difference between estimated and measured blood activity at 1-3 sampling points; at least two blood samples were needed to get a good IDIF (Hackett et al., 2013).
IDIF methods that are available in TPC
- Myocardial perfusion using [15O]H2O PET
- IDIF from abdominal aorta in [15O]H2O PET
- Correction of myocardial LV muscle and cavity TACs
Alternative methods to retrieve model input from image data
Model-based input function is based on the assumption that the input function is common to all tissue regions in the image, and can be solved from the data.
Population-based input function (PBIF) is based on the individual scaling of a tracer- and population-specific input curve (Zanotti-Fregonara et al, 2013). For example injected dose and subject mass can be used in scaling, or a venous blood sample.
- Blood and Plasma TAC
- Fitting PET input curves
- Blood sampling
- Metabolite correction
- Model-based input function
- Population-based input function
- PET image clustering
- Partial volume and spillover effects in cardiac PET
- Simulations in GitLab
Boellaard R, van Lingen A, Lammertsma AA. Experimental and clinical evaluation of iterative reconstruction (OSEM) in dynamic PET: quantitative characteristics and effects on kinetic modeling. J Nucl Med. 2001; 42(5): 808-817.
Brix G, Bellemann ME, Hauser H, Doll J. Recovery-Koeffizienten zur Quantifizierung der arteriellen Inputfunktion aus dynamischen PET-Messungen: experimentelle und theoretische Bestimmung. Nuklearmedizin 2002; 41: 184-190. doi: 10.1055/s-0038-1623894.
Chen K, Bandy D, Reiman E, Huang S-C, Lawson M, Feng D, Yun L, Palant A. Noninvasive quantification of the cerebral metabolic rate for glucose using positron emission tomography, 18F-fluoro-2-deoxyglucose, the Patlak method, and an image-derived input function. J Cereb Blood Flow Metab. 1998; 18(7): 716-723.
Christensen AN, Reichkendler MH, Larsen R, Auerbach P, Højgaard L, Nielsen HB, Ploug T, Stallknecht B, Holm S. Calibrated image-derived input functions for the determination of the metabolic uptake rate of glucose with [18F]-FDG. Nucl Med Commun. 2014; 35: 353-361. doi: 10.1097/MNM.0000000000000063.
Croteau E, Lavallée E, Labbe SM, Hubert L, Pifferi F, Rousseau JA, Cunnane SC, Carpentier AC, Lecomte R, Benard F. Image-derived input function in dynamic human PET/CT: methodology and validation with 11C-acetate and 18F-fluorothioheptadecanoic acid in muscle and 18F-fluorodeoxyglucose in brain. Eur J Nucl Med Mol Imaging 2010; 37: 1539-1550. doi: 10.1007/s00259-010-1443-z.
Fung EK, Carson RE. Cerebral blood flow with [15O]water PET studies using an image-derived input function and MR-defined carotid centerlines. Phys Med Biol. 2013; 58: 1903-1923. doi: 10.1088/0031-9155/58/6/1903.
Germano G, Chen BC, Huang S-C, Gambhir SS, Hoffman EJ, Phelps ME. Use of abdominal aorta for arterial input function determination in hepatic and renal PET studies. J Nucl Med. 1992;33:613-620.
Henze E, Huang S-C, Ratib O, Hoffman E, Phelps ME, Schelbert HR. Measurements of regional tissue and blood-pool radiotracer concentrations from serial tomographic images of the heart. J Nucl Med. 1983; 24: 987-996.
Hoekstra CJ, Hoekstra OS, Lammertsma AA. On the use of image derived input functions in oncological FDG PET studies. Eur J Nucl Med. 1999; 26(11): 1489-1492. doi: 10.1007/s002590050484.
Iguchi S, Hori Y, Moriguchi T, Morita N, Yamamoto A, Koshino K, Kawashima H, Zeniya T, Enmi J, Iida H. Verification of a semi-automated MRI-guided technique for non-invasive determination of the arterial input function in 15O-labeled gaseous PET. Nucl Instr Methods Phys Res A. 2013; 702: 111-113. doi: 10.1016/j.nima.2012.08.037.
Iida H, Rhodes CG, de Silva R, Araujo LI, Bloomfield P, Lammertsma AA, Jones T. 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; 33:1669-1677.
Lanz B, Poitry-Yamate C, Gruetter R. Image-derived input function from the vena cava for 18F-FDG PET studies in rats and mice. J Nucl Med. 2014; 55: 1380-1388. 10.2967/jnumed.113.127381.
Lüdemann L, Sreenivasa G, Michel R, Rosner C, Plotkin M, Felix R, Wust P, Amthauer H. Corrections of arterial input function for dynamic H215O PET to assess perfusion of pelvic tumours: arterial blood sampling versus image extraction. Phys Med Biol. 2006; 2883-2900. doi: 10.1088/0031-9155/51/11/014.
Lyoo CH, Zanotti-Fregonara P, Zoghbi SS, Liow J-S, Xu R, Pike VW, Zarate CA Jr, Fujita M, Innis RB. Image-derived input function derived from a supervised clustering algorithm: methodology and validation in a clinical protocol using [11C](R)-rolipram. PLoS ONE 2014; 9(2): e89101. doi: 10.1371/journal.pone.0089101.
Mourik JEM, Lubberink M, Schuitemaker A, Tolboom N, van Berckel BNM, Lammertsma AA, Boellaard R. Image-derived input functions for PET brain studies. Eur J Nucl Med Mol Imaging 2009; 36: 463-471. doi: 10.1007/s00259-008-0986-8.
Nelissen N, Warwick J, Dupont P (2012). Kinetic modelling in human brain imaging. In: Positron Emission Tomography - Current Clinical and Research Aspects, Dr. Chia-Hung Hsieh (Ed.), ISBN: 978-953-307-824-3, InTech. doi: 10.5772/30052.
Ohtake T, Kosaka N, Watanabe T, Yokoyama I, Moritan T, Masuo M, Iizuka M, Kozeni K, Momose T, Oku S, Nishikawa J, Sasaki Y, Iio M. Noninvasive method to obtain input function for measuring tissue glucose utilization of thoracic and abdominal organs. J Nucl Med. 1991; 32: 1432-1438.
Richard MA, Fouquet JP, Lebel R, Lepage M. MRI-guided derivation of the input function for PET kinetic modeling. PET Clin. 2016; 11: 193-202: doi: 10.1016/j.cpet.2015.09.003.
Schain M, Benjaminsson S, Varnäs K, Forsberg A, Halldin C, Lansner A, Farde L, Varrone A. Arterial input function derived from pairwise correlations between PET-image voxels. J Cereb Blood Flow Metab. 2013; 33: 1058-1065. doi: 10.1038/jcbfm.2013.47.
Su Y, Arbelaez AM, Benzinger TLS, Snyder AZ, Vlassenko AG, Mintun MA, Raichle ME. Noninvasive estimation of the arterial input function in positron emission tomography imaging of cerebral blood flow. J Cereb Blood Flow Metab. 2013; 33: 115-121. doi: 10.1038/jcbfm.2012.143.
van der Weerdt AP, Klein LJ, Boellaard R, Visser CA, Lammertsma AA. Image-derived input functions for determination of MRGlu in cardiac 18F-FDG PET scans. J Nucl Med. 2001; 42:1622-1629.
Zanotti-Fregonara P, Chen K, Liow J-S, Fujita M, Innis RB. Image-derived input function for brain PET studies: many challenges and few opportunities. J Cereb Blood Flow Metab. 2011; 31: 1986-1998. doi: 10.1038/jcbfm.2011.107.
Zanotti-Fregonara P, Hirvonen J, Lyoo CH, Zoghbi SS, Rallis-Frutos D, Huestis MA, Morse C, Pike VW, Innis RB. Population-based input function modeling for [18F]FMPEP-d2, an inverse agonist radioligand for cannabinoid CB1 receptors: validation in clinical studies. PLoS One. 2013; 8(4): e60231. doi: 10.1371/journal.pone.0060231.
Zheng X, Tian G, Huang S-C, Feng D. A hybrid clustering method for ROI delineation in small-animal dynamic PET images: application to the automatic estimation of FDG input functions. IEEE Trans Inform Technol Biomed. 2011; 15(2): 195-205. doi: 10.1109/TITB.2010.2087343.
Updated at: 2019-01-05
Created at: 2010-03-14
Written by: Vesa Oikonen