| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Basic Science Investigations |
PET Center, VU University Medical Center, Amsterdam, The Netherlands
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
50% was found for all SUV values. For nonsmoothed data with high noise levels this variation was primarily due to noise, whereas for smoothed data with low noise levels partial-volume effects were most important. In general, SUVs showed under- and overestimations of 50% and depended on all parameters studied. However, SUV ratios, used for response monitoring, were only slightly dependent of ROI definition but were still affected by noise and resolution. Conclusion: The poor accuracy of the SUV under various conditions may hamper its use for diagnosis, especially in multicenter trials. SUV ratios used to measure response to treatment, however, are less dependent on noise, image resolution, and ROI definition. Therefore, the SUV might be more suitable for response-monitoring purposes.
Key Words: standard uptake value • response monitoring • region of interest • image resolution • noise
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
The first step in determining the SUV is to derive the activity concentration (AC) in the tumor. Usually, the AC is obtained by placing a region of interest (ROI) over the tumor either visually (11), automatically using a threshold value (12–16), or using a fixed size (10). Unfortunately, various ROI methods are in use, making it difficult to compare different studies. Consequently, results obtained in one institution may not apply to results from other institutions. Other factors, such as reconstruction algorithm and filter, scanner sensitivity and scan duration, sinogram noise, and partial-volume effects, can also affect the accuracy of the measured AC (17–25).
The purpose of this study was to evaluate the effects of noise, image resolution, and ROI definition on the accuracy of measured SUVs. To this end, both simulation and experimental studies were performed, allowing us to separately investigate the effects of tumor size, tumor-to-background ratio (TBR), noise, image resolution, and ROI definitions on the accuracy of the measured AC. Furthermore, the effects on the accuracy of the observed relative changes in SUV were also investigated.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Phantom Study
An anthropomorphic thorax phantom (Data Spectrum) containing 2 lungs and a liver insert was used. The large background compartment (soft tissue) was filled with an 18F-FDG solution of 5 kBq/mL. Spheres, representing tumors of 8, 12, and 30 mm, were filled with an 18F-FDG solution of 20 kBq/mL (TBR = 4) and positioned in the mediastinum region of the phantom. Twenty-five 2-dimensional (2D) emission scans, each of 900 kilocounts, corresponding to the average number of counts observed in 5-min patient scans, were acquired using an ECAT HR+ scanner (CTI/Siemens) (26,27). By performing acquisitions terminated on the number of acquired counts, statistically equivalent sinograms were obtained, allowing for reproducibility assessment. A 5-min transmission scan was acquired for attenuation correction. Data were reconstructed using ordered-subset expectation maximization (OSEM) with 2 iterations and 16 subsets (18). Gaussion filters of 0- and 6-mm full width at half maximum (FWHM) were used to match the image resolutions used in the simulations.
Simulation Study I: SUV Accuracy and Relative Change
A mathematic phantom was derived from a 2D dynamic 18F-FDG PET scan of a patient. The patient data were summed from 15 to 45 min, and the resulting sinogram was reconstructed using OSEM with 2 iterations and 16 subsets and postsmoothed using a 6-mm FWHM gaussian filter. Subsequently, pixel values were scaled to an AC of 
5 kBq/mL, corresponding to the average soft-tissue AC in patient studies. In this mathematic phantom, tumors were simulated with spheres. Tumor sizes of 10-, 15-, 20-, and 30-mm diameter were used to cover the lower part of the clinically relevant range, as determination of the SUV is most challenging for small tumors. Tumors were located in the breast and lung. TBRs of 2, 4, and 8 (10, 20, and 40 kBq/mL) were applied. The size of the mathematic phantom corresponds to that of a patient of about 80 kg.
Next, noise-free sinograms were generated by forward projection of the image. These sinograms represented the number of true coincidences. Random and scattered coincidences were added to obtain prompts. Randoms were assumed to be distributed uniformly over the sinogram. Scattered coincidences were derived from forward projecting the difference between a scatter-corrected and a noncorrected image. Poisson noise was added to all sinograms. True coincidence sinograms were generated by subtracting the noisy random and scatter sinograms from the noisy prompts sinograms. Three noise levels were simulated corresponding to 2D data obtained for 3–5, 7–10, and 30–40 min or 0.75E+7, 1.5E+7, and 6.0E+7 noise equivalent counts (NEC) (28), respectively. For each combination of tumor size, TBR, and noise level, 50 noisy sinograms were generated to evaluate reproducibility and bias of the SUVs.
All sinograms were reconstructed using OSEM with 2 iterations and 16 subsets (ECAT version 7.2; CTI/Siemens). Image matrix sizes of 128 x 128 or 256 x 256, corresponding to pixel sizes of 5.12 x 5.12 and 2.56 x 2.56 mm, were used. Reconstructed images were postsmoothed using a gaussian filter such that image resolutions equaled 5- and 8-mm FWHM.
Finally, the phantom was adjusted by 25% to simulate patient weights of 60 and 100 kg, respectively. The number of acquired counts was adjusted using an empirically derived relation between patient weight and the counts-per-minute emission scan.
Simulation Study II: Noise and Bias
To assess the relations between SUV, tumor size, and noise, a simulation was performed using a uniform phantom containing an AC of 5 kBq/mL. Various noise levels, obtained by adding Poisson noise to the simulated sinograms, were applied, resulting in coefficients of variation (COVs) of 0%, 10%, 20%, 35%, and 50%. For each noise level, 100 simulations were performed. Spheric 3-dimensional (3D) ROIs of 11, 15, 21, 32, 39, and 50 mm in diameter were projected onto these images, and the maximum pixel value within each ROI was derived. The average maximum pixel value and its SE for each ROI size over the 100 simulations per noise level were calculated. These average values were normalized to true AC (5 kBq/mL).
Data Analysis
The same ROI methods were used for the phantom and the simulation study:
These ROIs cover the various types of ROIs that are in regular use (12–15,29). Manual definition of ROIs (30) was not attempted, because it was impossible (21,600 ROIs) for the present study. Moreover, other studies have shown that automatic definition of the ROI improves the interobserver reproducibility and accuracy of the measured AC (14). ROI0.5(Max+BG) was chosen because this ROI theoretically corresponds most closely to the actual tumor size. Furthermore, definition of ROI50 is not always possible for small tumors with low uptake, where ROI50 is lower than the background AC. ROI0.5(Max+BG) avoids this problem. ROI15x15 was included as a representative of a fixed-sized ROI (10).
Fifty AC values for the simulation study data and 25 for the phantom data were obtained for each combination of ROI, tumor size, TBR, tumor location, noise level, reconstruction matrix, and resolution. The accuracy of the observed AC was obtained from the average value over 50 simulated or 25 measured data. The recovery coefficient is defined here as the ratio between the observed AC and the true simulated AC. Note that the recovery coefficient may be larger than 1.0, indicating an overestimation of the actual AC.
The accuracy of measured relative changes, simulating treatment response, was investigated using the same simulation. Two situations were considered: (a) tumors of 10- and 30-mm diameter with a decrease in uptake, but no variation in size; and (b) tumors with TBR = 2 or 8 with a decrease in tumor size, but not in 18F-FDG uptake.
Definition of "Defaults"
An extensive amount of data was generated by the simulations and phantom studies. Therefore, unless stated otherwise, results presented here are limited to a default situation:
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
|
Simulation Study I: SUV Accuracy
Results obtained for breast tumors were almost identical to lung tumor data. For lung tumors, recovery coefficients were slightly lower due to a larger contrast with background activity. Therefore, results are presented for simulated breast tumors only.
Figures 2A–2D show recovery coefficients for data obtained with various noise levels and different image resolutions. For clarity, data based on maximum pixel value and ROI50 are presented, as intermediate results were obtained for other thresholds. Figure 2A shows that use of maximum pixel values resulted in large overestimations of the AC, which increased with higher noise levels. A similar variation with noise levels was found for ROI50 values (Fig. 2B), but these were more in agreement with the actual AC. For smoothed data (Figs. 2C and 2D), differences in recovery coefficients between various noise levels were much smaller, due to the noise-suppressing effect of smoothing. Substantial variations of recovery coefficient with tumor size were found for both smoothed and nonsmoothed data.
|
|
|
20% higher and 10% lower ACs were found at a 25% decrease or increase of patient weight. For smoothed data, differences are <5%.
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
In addition, the simulation focused on tumors located inthe thorax. When studies are performed at other body regions, somewhat different recovery coefficients can be expected due to differences in noise level resulting from differences in attenuation losses, random and scatter contribution. However, the variation of the SUV with resolution, noise, and ROI method will be similar, because noise, resolution, and ROI method are the underlying causes of the observed variations.
Accuracy of SUV
The observed recovery coefficients showed similar variation with sphere size under various conditions for all ROIs evaluated. There is, however, a difference for smoothed and unsmoothed data. In Table 3, values for the COV of background pixels are given for both smoothed and nonsmoothed data for each noise level. It can be deduced that image noise varies more with sinogram statistics for nonsmoothed data than for smoothed data. The relationship between image noise and sinogram statistics is consistent with those observed by Boellaard et al. (18). The results presented in Figure 2 indicate that the maximum pixel value increases both with ROI or tumor size and with noise level. Equivalently, maximum pixel values within an object increase with image noise level and with increasing object size. Similar results were observed by Falen et al. (21), who found SUV increases up to 70% with a higher number of iterations. Increasing the number of iterations not only improves convergence but also increases image noise. Therefore, images with higher noise show more positive bias for both the maximum pixel and the ROI value within a sphere (Fig. 7), and this bias increases with sphere size. This explains the large variation of maximum pixel value with object size for nonsmoothed data. Note that the variation of the maximum pixel value for nonsmoothed data is similar to that for smoothed data, indicating that partial-volume effects have only a minor contribution to the observed dependence on sphere size. Smoothing the data reduces effects of image noise but introduces a larger partial-volume effect. Consequently, variations of the maximum pixel value with sphere or tumor size are substantial for both smoothed and nonsmoothed images, but the underlying mechanisms are different. Because all isocontour-based ROI methods use the maximum pixel value as a reference, similar results were found using these ROIs (Figs. 1C and 1D).
|
The Nyquist principle requires the pixel size to be at least 2 times smaller than image resolution. For the simulated images with a resolution of 5-mm FWHM, a matrix size of 128 x 128 or a pixel size of 5 x 5 x 2.5 mm does not fulfill this criterion and will reduce image resolution (Table 1). Reducing the pixel size to 2.5 x 2.5 mm does no longer violate the Nyquist principle and will therefore not degrade image resolution. As expected, for smoothed data, with an image resolution of 8-mm FWHM, the effect of matrix or pixel size was much smaller.
Finally, patient weight has an effect on the accuracy of the measured AC. Variation in weight causes variation in sinogram counts or noise (Table 2), which causes bias. When data are sufficiently smoothed, the effects of noise are minimized at the cost of an increased partial-volume effect. A useful strategy to overcome this problem might be to adjust the dose to patient weight.
Response Monitoring
On average, the measured response was almost independent of the ROI method when the tumor size was kept constant. The good agreement of the measured response among all ROIs can be understood from the data presented in Figures 1C and 1D, which illustrate the strong correlation of ROI values over all noisy realizations. Therefore, the reproducibility and accuracy of the AC of the maximum pixel value in fact determines the reproducibility and accuracy of the AC measured with an isocontour ROI. Consequently, similar responses were measured with all isocontour-based ROIs. A slightly lower accuracy and very poor reproducibility were observed for ROI15x15 data because this region is too large for the small tumor sizes investigated in this study.
The measured response for tumors that showed a large variation in size strongly depended on tumor size, which is a logical consequence of the data presented in Figures 1A and 1B and in Figures 7A and 7B. Smoothing the data reduced, but did not remove, the differences among the ROIs. As mentioned previously, both noise-induced bias for nonsmoothed data and partial-volume effects for smoothed data are the underlying causes of the observed effects.
The (almost) independence of the measured response on noise, resolution, and ROI method is explained by the fact that most factors cancel out when calculating SUV ratios, such as bias due to noise; absolute quantification due to the applied ROI method; patient weight (= noise); image resolution. When the tumor size does not vary much, partial-volume effects will also cancel out (at least partially). However, more accurate results can be expected when accurate partial-volume corrections can be applied. Note, however, that smoothing improves the reproducibility of the observed responses.
Considerations for Use of SUV in Clinical Practice
From the present study it can be concluded that the SUV obtained under specific conditions may not be compared directly with those obtained, reconstructed, or analyzed under other conditions (31). Consistency of data acquisition and analysis protocols is therefore required. This can easily be adhered to within one institution, explaining the successful use of the SUV in differentiating benign from malignant tumors. However, SUV measures will probably vary strongly among institutions. Therefore, SUV threshold values used to differentiate between malignant and benign lesions should not be taken from the literature and used without validating the appropriateness. Moreover, another important observation is made by Thie et al. (32), who found that SUVs show a lognormal rather than a gaussian distribution, which will require a new review of applied SUV thresholds. Finally, standardization of imaging and analysis protocols is required for multicenter studies. Standardization is required, for example, for acquisition mode; average NEC (combination of scanner sensitivity and scan durations); reconstruction method and image resolution; interval between 18F-FDG administration and scanning; partial-volume corrections; head or feet at first bed position; ROI method; emptying of bladder; SUV calculation method. Any remaining differences may be determined and corrected for by including a standardized phantom experiment—for example, using anthropomorphic phantoms (Data Spectrum).
Another application of the SUV is to measure the treatment response of a tumor (5,6,8,10). Assuming consistent data processing for multiple scans of 1 patient, the measured response might be less dependent on noise, image resolution, and ROI definition (Figs. 4–6). Minimizing noise levels by smoothing reduces bias and seems to have a minimal effect on the accuracy of the measured response without the need for partial-volume correction algorithms. Some partial-volume effects were observed, however, when the tumor volume changed drastically. This can only be solved by the introduction of partial-volume correction methods. These variations in tumor size, however, were most likely an extreme situation in relation to changes seen in clinical practice. Because of the relative independence of the SUV response with respect to several study parameters, multicenter response-monitoring studies should be feasible.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
For correspondence contact: Ronald Boellaard, PhD, PET Center, VU University Medical Center, De Boelelaan 1117, 1081 HV Amsterdam, The Netherlands.
E-mail: r.boellaard{at}vumc.nl
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Related articles in JNM:
This article has been cited by other articles:
|
|
 |
|
  R. L. Wahl, H. Jacene, Y. Kasamon, and M. A. Lodge From RECIST to PERCIST: Evolving Considerations for PET Response Criteria in Solid Tumors J. Nucl. Med., May 1, 2009; 50(Suppl_1): 122S - 150S. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Boellaard Standards for PET Image Acquisition and Quantitative Data Analysis J. Nucl. Med., May 1, 2009; 50(Suppl_1): 11S - 20S. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Itti, C. Lin, J. Dupuis, G. Paone, D. Capacchione, A. Rahmouni, C. Haioun, and M. Meignan Prognostic Value of Interim 18F-FDG PET in Patients with Diffuse Large B-Cell Lymphoma: SUV-Based Assessment at 4 Cycles of Chemotherapy J. Nucl. Med., April 1, 2009; 50(4): 527 - 533. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. W. L. Boland, M. A. Blake, N. S. Holalkere, and P. F. Hahn PET/CT for the Characterization of Adrenal Masses in Patients with Cancer: Qualitative Versus Quantitative Accuracy in 150 Consecutive Patients Am. J. Roentgenol., April 1, 2009; 192(4): 956 - 962. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Nakajo, S. Jinnouchi, Y. Tashiro, H. Shirahama, E. Sato, C. Koriyama, and M. Nakajo Effect of Clinicopathologic Factors on Visibility of Colorectal Polyps with FDG PET Am. J. Roentgenol., March 1, 2009; 192(3): 754 - 760. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. M. Park, J. M. Goo, H. J. Lee, M. A Kim, C. H. Lee, and M.-J. Kang Tumors in the Tracheobronchial Tree: CT and FDG PET Features RadioGraphics, January 1, 2009; 29(1): 55 - 71. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. R. Benz, V. Evilevitch, M. S. Allen-Auerbach, F. C. Eilber, M. E. Phelps, J. Czernin, and W. A. Weber Treatment Monitoring by 18F-FDG PET/CT in Patients with Sarcomas: Interobserver Variability of Quantitative Parameters in Treatment-Induced Changes in Histopathologically Responding and Nonresponding Tumors J. Nucl. Med., July 1, 2008; 49(7): 1038 - 1046. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Lin, E. Itti, C. Haioun, Y. Petegnief, A. Luciani, J. Dupuis, G. Paone, J.-N. Talbot, A. Rahmouni, and M. Meignan Early 18F-FDG PET for Prediction of Prognosis in Patients with Diffuse Large B-Cell Lymphoma: SUV-Based Assessment Versus Visual Analysis J. Nucl. Med., October 1, 2007; 48(10): 1626 - 1632. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. B. Dolovich and D. P. Schuster Positron Emission Tomography and Computed Tomography versus Positron Emission Tomography Computed Tomography: Tools for Imaging the Lung Proceedings of the ATS, August 1, 2007; 4(4): 328 - 333. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Soret, S. L. Bacharach, and I. Buvat Partial-Volume Effect in PET Tumor Imaging J. Nucl. Med., June 1, 2007; 48(6): 932 - 945. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B.-K. Teo, Y. Seo, S. L. Bacharach, J. A. Carrasquillo, S. K. Libutti, H. Shukla, B. H. Hasegawa, R. A. Hawkins, and B. L. Franc Partial-Volume Correction in PET: Validation of an Iterative Postreconstruction Method with Phantom and Patient Data J. Nucl. Med., May 1, 2007; 48(5): 802 - 810. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. Bastiaannet, O. S. Hoekstra, W. J. G. Oyen, P. L. Jager, T. Wobbes, and H. J. Hoekstra Level of Fluorodeoxyglucose Uptake Predicts Risk for Recurrence in Melanoma Patients Presenting With Lymph Node Metastases Ann. Surg. Oncol., July 1, 2006; 13(7): 919 - 926. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. K. Shankar, J. M. Hoffman, S. Bacharach, M. M. Graham, J. Karp, A. A. Lammertsma, S. Larson, D. A. Mankoff, B. A. Siegel, A. Van den Abbeele, et al. Consensus Recommendations for the Use of 18F-FDG PET as an Indicator of Therapeutic Response in Patients in National Cancer Institute Trials J. Nucl. Med., June 1, 2006; 47(6): 1059 - 1066. [Full Text] [PDF]
|
|
|
|
|
|
Y. Hashimoto, T. Tsujikawa, C. Kondo, M. Maki, M. Momose, A. Nagai, T. Ohnuki, T. Nishikawa, and K. Kusakabe Accuracy of PET for Diagnosis of Solid Pulmonary Lesions with 18F-FDG Uptake Below the Standardized Uptake Value of 2.5 J. Nucl. Med., March 1, 2006; 47(3): 426 - 431. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. Bunyaviroch and R. E. Coleman PET Evaluation of Lung Cancer J. Nucl. Med., March 1, 2006; 47(3): 451 - 469. [Full Text] [PDF]
|
|
|
|
|
|
V. Kersemans, B. Cornelissen, K. Bacher, K. Kersemans, H. Thierens, R. A. Dierckx, B. De Spiegeleer, G. Slegers, and J. Mertens In Vivo Evaluation and Dosimetry of 123I-2-Iodo-D-Phenylalanine, a New Potential Tumor-Specific Tracer for SPECT, in an R1M Rhabdomyosarcoma Athymic Mouse Model J. Nucl. Med., December 1, 2005; 46(12): 2104 - 2111. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. J. Hoekstra, S. G. Stroobants, E. F. Smit, J. Vansteenkiste, H. van Tinteren, P. E. Postmus, R. P. Golding, B. Biesma, F. J.H.M. Schramel, N. van Zandwijk, et al. Prognostic Relevance of Response Evaluation Using [18F]-2-Fluoro-2-Deoxy-D-Glucose Positron Emission Tomography in Patients With Locally Advanced Non-Small-Cell Lung Cancer J. Clin. Oncol., November 20, 2005; 23(33): 8362 - 8370. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W A Weber PET for response assessment in oncology: radiotherapy and chemotherapy Br. J. Radiol., November 1, 2005; Supplement_28(1): 42 - 49. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. J. Herder, H. van Tinteren, R. P. Golding, P. J. Kostense, E. F. Comans, E. F. Smit, and O. S. Hoekstra Clinical Prediction Model To Characterize Pulmonary Nodules: Validation and Added Value of 18F-Fluorodeoxyglucose Positron Emission Tomography Chest, October 1, 2005; 128(4): 2490 - 2496. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Pasic, H. A. Brokx, E. F. Comans, G. J. Herder, E. K. Risse, O. S. Hoekstra, P. E. Postmus, and T. G. Sutedja Detection and Staging of Preinvasive Lesions and Occult Lung Cancer in the Central Airways with 18F-Fluorodeoxyglucose Positron Emission Tomography: A Pilot Study Clin. Cancer Res., September 1, 2005; 11(17): 6186 - 6189. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. A. Weber Use of PET for Monitoring Cancer Therapy and for Predicting Outcome J. Nucl. Med., June 1, 2005; 46(6): 983 - 995. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. A. Thie Understanding the Standardized Uptake Value, Its Methods, and Implications for Usage J. Nucl. Med., September 1, 2004; 45(9): 1431 - 1434. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY | THE JOURNAL OF NUCLEAR MEDICINE |
(A and B only) = maximum pixel value; 
= ROI70; 
= ROI0.5(BG+MAX); 
= ROI50; • = ROI15x15.
