A new partial volume correction method for dynamic FDG images of heterogeneous tumor using factor analysis and stepwise procedure
Kenji HirataK.P. WongWei ShaYe HuKeisuke S. IwamotoMoses Q. WilksDavid StoutWilliam H. McBrideNagara TamakiSung-Cheng Huang
0
Citation
0
Reference
20
Related Paper
Abstract:
2068 Objectives Development of tumor necrosis is often accompanied by metabolic heterogeneity. Partial volume correction (PVC) of FDG images of heterogeneous tumor that involves unknown irregular shape and non-uniform concentrations is difficult. We propose a new approach that integrated PVC with factor analysis that is applicable to dynamic FDG PET images of tumors. Methods A 60-min dynamic FDG PET was performed on 11 SCID mice with implanted U87 glioblastoma. Eight out of 11 that had tumor necrosis were investigated. Factor analysis was applied to the dynamic tumor images with up to 4 factors. PVC was performed on the factor image of the tumor component as follows. A segment boundary was determined using a threshold t1(%) of maximum, and the segment image was smoothed using the scanner’s resolution. The smoothed image was then subtracted from the original factor image. The procedure was repeated 3 more times with different thresholds (t2, t3, and t4(%)), which were selected for each animal to minimize the sum of square of the residual image. Results Regardless of the number of factors used in the factor analysis, the FDG kinetics of all tumor tissues were represented by only a single factor (i.e., no other factor images showed tumor structure). The kinetic characteristics of the corresponding factor were close to the kinetics obtained from the tumor ROI. The smoothed PVC images were visually similar to the original tumor factor images. Quantitatively, relative difference between the smoothed PVC image and the original factor image was averaged 9.2±1.1% (range: 8.2 to 11.3%) for all 8 tumors studied. The total activity of the PVC images was within 5% of that of the original factor image. Conclusions This new method using factor analysis followed by stepwise correction procedure successfully converted dynamic tumor FDG images to PVC images of tumor FDG kinetics. This approach is expected to improve quantitation of FDG uptake in tumors with necrosis for longitudinal tumor progression studies.Keywords:
Shape factor
Partial volume
Cite
The standardized uptake value (SUV) is the most commonly used parameter to quantify the intensity of radiotracer uptake in tumors. Previous studies suggested that measurements of (18)F-FDG accumulation in tissue might be affected by the image reconstruction method, but the clinical relevance of these findings has not been assessed.Phantom studies were performed and clinical whole-body (18)F-FDG PET images of 85 cancer patients were analyzed. All images were reconstructed using either filtered backprojection (FBP) with measured attenuation correction (MAC) or iterative reconstruction (IR) with segmented attenuation correction (SAC). In a subset of 15 patients, images were reconstructed using all 4 combinations of IR+SAC, IR+MAC, FBP+SAC, and FBP+MAC. For phantom studies, a sphere containing (18)F-FDG was placed in a water-filled cylinder and the activity concentration of that sphere was measured in FBP and IR reconstructed images using all 4 combinations. Clinical studies were displayed simultaneously and identical regions of interest (ROIs, 50 pixels) were placed in liver, urinary bladder, and tumor tissue in both image sets. SUV max (maximal counts per pixel in ROI) and SUV avg (average counts per pixel) were measured.In phantom studies, measurements from FBP images underestimated the true activity concentration to a greater degree than those from IR images (20% vs. 5% underestimation). In patient studies, SUV derived from FBP images were consistently lower than those from IR images in both normal and tumor tissue: Tumor SUV max with IR+SAC was 9.6 +/- 4.5, with IR+MAC it was 7.7 +/- 3.5, with FBP+MAC it was 6.9 +/- 3.0, and with FBP+SAC it was 8.6 +/- 4.1 (all P < 0.01 vs. IR+SAC). Compared with IR+SAC, SUV from FBP+MAC images were 25%-30% lower. Similar discrepancies were noted for liver and bladder. Discrepancies between measurements became more apparent with increasing (18)F-FDG concentration in tissue.SUV measurements in whole-body PET studies are affected by the applied methods for both image reconstruction and attenuation correction. This should be considered when serial PET studies are done in cancer patients. Moreover, if SUV is used for tissue characterization, different cutoff values should be applied, depending on the chosen method for image reconstruction and attenuation correction.
Standardized uptake value
Correction for attenuation
Cite
Citations (87)
Tumor standardized uptake values (SUVs) vary with the interval between 18F-FDG injection and image acquisition. This paper presents a simple method using a single reference point to make appropriate time corrections for tumor SUVs.The reference point method was algebraically deduced from observations made by Beaulieu et al., who found that tumor SUVs behaved linearly over time (∼30 to 75 min after 18F-FDG injection). Eighteen patients with breast cancer were dynamically examined with PET/CT (∼60 and 80 min after 18F-FDG injection). Maximum SUV was calculated by applying 2 different iterative reconstruction methods (high-definition reconstruction and attenuation-weighted ordered-subsets expectation maximization). Reference points for time corrections were given, and errors for corrections obtained with the reference point method were calculated.Variations in the reconstruction algorithm strongly influenced the coordinates of the reference point. Time corrections using the reference point method were more accurate at higher tumor SUVs (>8 at high-definition reconstruction and>6 at attenuation-weighted ordered-subsets expectation maximization) than at lower ones.A common origin of tumor SUVs over time exists in breast cancer. In combination with the linear behavior of tumor SUVs between approximately 30 and 80 min, such a reference point allows for straightforward time corrections of tumor SUVs. Parameters for image reconstruction must be considered because they influence the coordinates of the reference point.
Time point
Correction for attenuation
Cite
Citations (9)
Cite
Citations (0)
Parametric imaging has been shown to provide better quantitation physiologically than SUV imaging in PET. With the increased sensitivity from a recently developed total-body PET scanner, whole-body scans with higher temporal resolution become possible for dynamic analysis and parametric imaging. In this paper, we focus on deriving the parameter k1 using compartmental modeling and on developing a method to acquire whole-body 18F-FDG PET parametric images using only the first 90 s of the postinjection scan data with the total-body PET system. Methods: Dynamic projections were acquired with a time interval of 1 s for the first 30 s and a time interval of 2 s for the following minute. Image-derived input functions were acquired from the reconstructed dynamic sequences in the ascending aorta. A 1-tissue-compartment model with 4 parameters (k1, k2, blood fraction, and delay time) was used. A maximum-likelihood-based estimation method was developed with the 1-tissue-compartment model solution. The accuracy of the acquired parameters was compared with the ones estimated using a 2-tissue-compartment irreversible model with 1-h-long data. Results: All 4 parametric images were successfully calculated using data from 2 volunteers. By comparing the time-activity curves acquired from the volumes of interest, we showed that the parameters estimated using our method were able to predict the time-activity curves of the early dynamics of 18F-FDG in different organs. The delay-time effects for different organs were also clearly visible in the reconstructed delay-time image with delay variations of as large as 40 s. The estimated parameters using both 90-s data and 1-h data agreed well for k1 and blood fraction, whereas a large difference in k2 was found between the 90-s and 1-h data, suggesting k2 cannot be reliably estimated from the 90-s scan. Conclusion: We have shown that with total-body PET and the increased sensitivity, it is possible to estimate parametric images based on the very early dynamics after 18F-FDG injection. The estimated k1 might potentially be used clinically as an indicator for identifying abnormalities.
Whole body imaging
Dynamic imaging
Region of interest
Temporal resolution
PET Imaging
Cite
Citations (74)
A new method for quantitating total lesion glucose metabolic changes in serial tumor FDG PET studies
Accurate quantitative FDG PET studies have the potential for important applications in clinical oncology for monitoring therapy induced changes in tumor glycolytic rates. Due to a number of technical problems that complicate the use of quantitative PET tumor imaging, methods which can maximize the accuracy and precision of such measurements are advantageous. In this study, we developed and evaluated a method for reducing the errors caused by the conventional single plane, single ROI analysis in parametric images generated from pixel by pixel Patlak graphic analysis (PGA) in FDG PET studies of melanoma patients. We compared this new method to the conventional ROI method. The new processing method involves (1) generating the correlation coefficient (r) constrained Patlak parametric images from dynamic PET data; (2) summing up all the planes which cover the lesion; (3) defining a single ROI which covers the whole lesion in the summing image and determining the total lesion glucose metabolic index (K{sub T}, ml/min/lesion). Although only a single ROI was defined on the summing image, the glucose metabolic index obtained showed negligible difference (<1%) compared to those obtained from multiple ROIs on multiple planes of unconstrained parametric images. When the dynamic PET images were rotated and translatedmore » to simulate different patient positionings between scans at different times, the results obtained from the new method showed negligible difference (<2%). In summary, we present a simple but reliable method to quantitatively monitor the total lesion glucose metabolic changes during tumor growth. The method has several advantages over the conventional single ROI, single plane evaluation: (1) less sensitive to the ROI definition; (2) smaller intra- and inter-observer variations and (3) not requiring image registrations of serial scan data.« less
Region of interest
Cite
Citations (0)
Objective: To develop an auto-segmentation method for delineating the biological tumor volume of nasopharyngeal carcinoma from positron emission tomography images. Methods: A phantom consisting of a water tank with fixed background fluorodeoxyglucose [18F-FDG] activity and spheres with diameters ranging from 2.1 to 5 cm with varying activities of FDG were used to simulate tumors of different sizes and FDG uptake. The phantom was scanned with a PET/CT scan at different sphere to background intensity ratios. An optimum fixed percentage threshold (FT) approach and a signal-to-background ratio (SBR) approach were developed to estimate the true size of the spheres from the PET images. Both approaches were further evaluated in patient images for validation. Twenty-two patients with NPC from stage T1 to T4 were included. The PET based biological tumor volumes (BTV) were delineated with both FT (BTVFT) and SBR (BTVSBR) approaches and compared with the gross tumor volume localized from MRI (GTVMR). The mean volumes of BTVFT and BTVSBR were compared and the degree of overlap between GTVMR and both BTVs was evaluated. Paired t-tests were used for statistical analysis. Results: The optimal FT value was 36.5% of maximal intensity, and SBR approach was represented by an inverse linear regression model. The estimated volume of spheres segmented by both approaches shows no significant difference from the true volume of spheres (p > 0.05), but the average absolute errors were smaller from SBR approach than FT approach (p = 0.008). GTVMR was larger than both BTVFT (p = 0.003) and BTVSBR (p SBR with GTVMR is significantly larger than with BTVFT (0.52 and 0.42 respectively, p MR for radiotherapy planning purpose.
Partial volume
Intensity
Standardized uptake value
Fluorodeoxyglucose
Cite
Citations (2)
1512 Objectives Respiration causes mismatch between CT and PET images, affecting PET quantification. We have recently developed an image-based method to correct for known errors in the attenuation factors. The method does not need re-reconstruction and is fast ( Methods Respiratory-gated CT and FDG-PET data were acquired on 5 patients at San Raffaele Hospital, Milan using a GE DSTE PET/CT scanner and Varian RPM tracking device. CT and PET data were gated into 6 matching phases. Each PET gate was reconstructed using each CT gate, introducing mismatch in 5 images for each PET gate. These images were corrected based on the difference of the mismatched and matched attenuation images. 9 lesions were segmented on each image, and meanSUV and volume was computed. The effect of attenuation mismatch was analyzed with the following metrics: 1) for every PET gate, the relative Root Mean Square Error (RRMSE) of the 5 mismatched data sets was computed and then averaged over all 6 PET gates; 2) the maximum relative error (maxRE). Results Segmented lesion volumes were 1-4cc, while maximum lesion displacement ranged between 6-12 mm. Conclusions Using CT data of a different respiration stage for attenuation correction affects quantification of SPNs. Changes in SUV values depend on the amount of motion and the surrounding tissue. The image-based correction decreased average variability in meanSUV from 7% to 3%. As the method is fast, it shows great potential for interactive correction of PET images for misalignment with CT.
Correction for attenuation
PET-CT
Hounsfield scale
Cite
Citations (0)
2243 Objectives Radioembolization using Yittrium-90 (Y-90) microspheres treatment for unresectable hepatic tumor requires image analysis using Tc-99m macro aggregated albumin (MAA) administered through a hepatic artery. The purpose of this research is to determine if static image analysis compared to SPECT analysis significantly affects the calculated tumor to normal ratios (TN). Methods Forty- seven patients with metastatic hepatic tumor were administered approximately 148 MBq of MAA. Following administration, static and SPECT images were obtained. TN were calculated using two different methods. First method used count data from regions of interest (ROI) drawn on SPECT images and was calculated as the counts per voxel for an iso-contour ROI based on the FWHM from the tumor divided by the counts per voxel for adjacent normal tissue. Second method used count data from an iso-contour ROI drawn on static images and was calculated as the average counts from tumor tissue divided by the average counts from adjacent normal tissue. The two methods were compared using Student’s T-test. Results Average TN from the method using the SPECT data was 5.1 ± 4.9 (range of 0.8 and 22.3). The average TN from static image analysis was 2.5 ± 0.8 (range of 1.2 and 4.3). The difference is a statistically significant (p-value = 0.0005). Conclusions TN range from SPECT analysis is wider when compared to the TN from static image analysis suggesting that the SPECT method may be more representative of the hetero-distribution of the microspheres within tumorous tissue. With the use of SPECT imaging, a more representative distribution of microspheres can lead to a better understanding of the effective treatment dose using Y-90 microspheres treatment for metastatic hepatic tumors.
Region of interest
Spect imaging
Cite
Citations (1)
1422 Objectives Image derived blood input functions of a mouse model for tracer kinetic analysis are confounded by partial volume effects, especially for FDG (myocardial uptake). Furthermore, tumors are often treated as homogeneous entity when modeling tracer kinetics. Here, both, FDG blood input time-activity curves (TACs) and tumor TACs, are extracted from a standard murine tumor model (EMT-6, BALB/C mice) using a novel non-negative matrix factorization (NMF) technique. Methods Dynamic FDG studies were acquired for 60 minutes (10x2sec, 8x5sec, 6x10sec, 6x20sec, 8x1min, 10x2min, 5x5min) on a small animal PET scanner (Inveon, Siemens, Knoxville, Tn). PET scans commenced right before FDG injection into the tail vein. The list mode data was sorted into sinograms and reconstructed (iterations of OSEM 3D followed by 18 MAP iterations, zoom x2). Masks were placed to outline the heart and tumor region (Rover software, ABX, Dresden, Germany). The time-activity series of all voxels contained in each mask served as input for the NMF algorithm. Results TACs of right and left ventricle (RV and LV), myocardium, and tumor were returned the NMF algorithm as factor curves. Their physical location was confirmed by associated factor volumes. Spillover between the myocardium and the ventricles was greatly reduced (Fig. 2(a)-(c)). The tumor volume was decomposed into three factor volumes representing vasculature, regions with steady increase (type-I) and regions with fast tracer uptake and gradual wash out (type-II), with maximum uptake of 62.2, 20.7, and 10.8 %id/g, respectively. Shape of the vasculature TAC within the tumor ROI was very similar to the cardiac blood pool TAC, as expected. Conclusions The proposed NMF technique decomposed the dynamic PET data into factor volumes and factor curves. It removed cardiac partial volume effects and revealed regional varying uptake characteristics in tumor. These provide the basis for modeling the tracer pharmacokinetics within the tumor.
Non-negative Matrix Factorization
Partial volume
Cite
Citations (1)
127 Objectives The Gjedde-Patlak plot (GP plot) is commonly used in quantitative dynamic FDG PET studies. The uptake rate constant Ki (mL/min/100 mL) can be theoretically estimated by the GP plot without knowledge of the full time course of tissue tracer kinetics, and this serves as a basis for the use of the GP plot to quantify PET scans with multi-bed positions. Our objective was to investigate the feasibility of parametric imaging for voxel-wise quantification of multi-bed dynamic FDG PET. Methods Twelve thoracic oncology FDG PET studies were performed on a GE Advance scanner at a single bed position. Dynamic images (256x256x35, voxel size 2x2x4.25 mm3) of 19 frames over 60 min (6x10s,3x20s,2x90s,2x150s,2x300s,4x600s) were reconstructed using an OSEM algorithm. To obtain the blood input function for the GP plot, ROIs of the left ventricle (LV) were drawn manually on the first 2 min images. The radioactivity distribution was used to calculate the center of mass of LV (mcLV). A volume (5x5x3) centered at mcLV was then copied to dynamic images to generate a time activity curve to be used for the input function. The GP plot with t* ≥ 10 min was used to generate Ki images. A 2 bed positions protocol was used to extract the samples from the continuous 10 to 60 min single-bed dynamic PET data without replacement. The SUV and Ki images calculated from the 10 to 60 min single-bed scans were used as reference for comparison. Results The Ki images generated from the samples of the 2 bed positions protocol were comparable to those generated from the standard continuous single-bed position data acquisition protocol. The Ki images showed higher contrast than SUV images to distinguish tumors from normal tissue. Conclusions The GP plot is a appropriate graphical method to generate reliable Ki images for voxel-wsie quantification of clinical dynamic FDG PET images obtained at multi-bed positions. Work in progress includes optimization of the data acquisition protocol, 4D image reconstruction, and the parametric imaging algorithm. Research Support KA24 DA00042(DFW)
Plot (graphics)
Gamma camera
Cite
Citations (2)