Evaluation of PET List Data-Driven Gated Motion Correction Technique Applied in Lung Tumors

Positron emission tomography/computed tomography (PET/CT) is an important tool for tumor staging or treatment response evaluation, especially for lung tumors. However, the captured static PET image could be blurry due to patients' free breathing, resulting in decreased image quality and incorrect quantitative values. This study aimed to evaluate whether the Q.Static scan mode with the novel PET list data-driven gated (DDG) technique decreases the lesion blurring problems in the PET/CT images of patients with lung cancer. Data of 194 patients with lung tumors were retrospectively reviewed. DDG Q.Static scan mode was set up in three beds to cover the whole chest and the upper abdomen in the routine PET/CT scans and was activated automatically when sensing significant respiratory motion. Routine reconstruction algorithm was applied for data analysis. Only the lesions in the motion-corrected areas were measured and calculated for statistics. Among the 194 patients, 124 had at least one bed that activated the DDG Q.Static procedure. However, only 49 out of the 124 patients showed lesions in their activated beds. Compared with the non-corrected data, the DDG Q.Static data showed improved accuracy with increased SUVmax and SUVmean of 8.52% (9.20 ± 5.42 to 9.74 ± 5.42) and 8.65% (6.11 ± 3.68 to 6.48 ± 3.68), respectively. In addition, metabolic tumor volume was reduced from 6.54 ± 8.58 to 5.55 ± 7.33 (14.79% reduction). For subjective image quality, the DDG Q.Static data scored higher than the non-corrected data. This study showed that the quantitative values and image quality were improved after the correction. Therefore, the DDG Q.Static technique is an effective method to correct motion artifacts in PET/CT scans.


Introduction
Positron emission tomography/computed tomography (PET/ CT) with 18F-fluoro-2-deoxyglucose (18F-FDG) offers simultaneous tumor metabolism image and anatomic image and is considered an accurate modality for evaluating lung cancer. It can effectively assess the malignancy of solitary pulmonary nodules and metastatic lesions. Owing to its excellent detection abilities, PET/CT scan has become an integral part of lung cancer staging [1][2][3][4]. However, static PET acquisitions take a few minutes for each axial field of view. As a result, the patients' free breathing causes PET image blurring, especially for tumors in the thorax and abdomen. These respiratory motion artifacts degrade PET image quality and potentially result in decreased lesion detectability, underestimated measured standard uptake value (SUV), and overestimated measured metabolic tumor volume (MTV), all of which influence image interpretation [5][6][7][8].
Many motion correction methods, e.g., elastic chest belt, real-time position management system, respiratory temperature sensor, and spirometer, could mitigate these artifacts by utilizing an external device to detect respiratory information [9][10][11][12]. However, setting up these external gating devices is time consuming, inefficient, and increases technologists' radiation exposure.
A deviceless data-driven gated (DDG) technique called Motionfree® (GE Healthcare, Waukesha, WI) is an algorithm based on principal component analysis and determines the magnitude of respiratory impact to the data. This method analyzes the coincidence data and converts them to respiratory waveforms without requiring any external devices [13][14][15]. Q.Static (GE Healthcare, Waukesha, WI) scan mode is a phase-based gating correction method that extracts a fraction of PET data from the end-expiration smooth parts of breathing cycles. By determining the relatively stable breathing period, the Q.Static reduces respiratory motion artifacts and generates a corrected image. However, executing the Q.Static scan mode requires an external device to detect respiratory information.
DDG Q.Static combines Q.Static and Motionfree to provide an easy and deviceless way to acquire end-expiration phase data. The user only needs to set the coverage for desired correction areas. DDG Qstatic is automatically activated and corrects the image when the respiratory motion is significant. This study aimed to assess whether the DDG Q.Static technique mitigates PET respiratory artifacts and improves PET image quality and quantitation accuracy in lung cancer cases.

Data Acquisition
This retrospective study was conducted from October 2019 to February 2020 in Taipei Veterans General Hospital. Data of patients with confirmed primary lung cancer and arranged whole-body PET/CT scans by clinical physicians were reviewed. Lung metastatic lesions cases or cases with too many lesions to measure were excluded. A total of 194 patients consisting of 98 men and 96 women with a mean age of 65 ± 13 years (range 32-90 years) and a mean BMI of 23.7 ± 3.6 kg/m 2 (range 16.9-33.3) were eligible according to the criteria. Data collection and analyses were approved by the Institutional Review Board.
The patients were all injected with 3.7 MBq/kg ± 10% of 18F-FDG after fasting for at least 4 h and reaching a serum glucose concentration of less than 200 mg/dl. At 60 min after radiopharmaceutical injection, PET/CT scans were obtained with Discovery MI DR (GE Healthcare, Waukesha, WI, USA). A clinical whole-body PET/CT protocol was used for each patient to acquire from the top of the skull to the middle of the thigh. A free-breathing CT was acquired first under the following parameters: tube voltage, 120 kVp; effective tube current, smart mA (range from 30 to 130 mA); pitch of 0.984:1; gantry rotation time 0.5 s; and 3.75 mm slice thickness. No intravenous contrast agents were used.
After CT scanning, static PET was acquired with 2 min per bed position using the list-mode and time of flight technique.
All the patients were requested to breathe steadily and regularly during the whole scan. The DDG Q.Static technique was arranged to detect respiratory motion from beds 3 to 5 that covered the whole thorax and upper abdomen (Fig. 1a). First, the DDG Q.Static would determine whether it needs to be activated. It analyzed the PET coincidence data and calculated a unitless ratio, R, which presents the strength of the signal being respiratory-like. The R value was calculated in real-time. The manufacturer suggested to set the R value of 15.0 as a threshold to assess whether the respiratory motion is significant. If the calculated R value exceeded the threshold, then the DDG Q.Static acquisition mode was activated. The PET acquired phase of DDG Q.Static was the preset as 30% offset and 50% width, which is the stable phase of the breathing cycle. The total acquisition time of the DDG Q.Static-activated bed was increased to achieve a consistent signal. The settings of phase offset and phase width were also suggested by the manufacturer (Fig. 1b). Then we recorded the bed whose R value was over the threshold for statistics.

Image Reconstruction and Analysis
Two series of images were reconstructed: DDG Q.Staticcorrected image (QSC) and non-corrected image (NC). Both images were subjected to the same CT-based attenuation correction and PET reconstruction algorithm, 3D VUE Point FX (ordered-subset expectation maximization algorithm with 2 iterations, 24 subsets, and Gauss-filtered to a transaxial resolution of 5 mm at full-width at halfmaximum (FWHM)) and Sharp IR (PSF model), and were reconstructed with a 256 × 256 matrix and 3.27 mm slice thickness. The CT and PET fusion images were produced and analyzed with AW server 3.2 (GE Healthcare, Waukesha, WI).
In QSC and NC, the 3D volume of interest (VOI) in the same size was drawn for the lung lesions, and the following were measured: maximum standardized uptake value (SUV max ), mean standardized uptake value (SUV mean ), and metabolic tumor volume (MTV). The VOI was semiautomatically drawn with an iso-contour threshold of 50% SUV max to reduce artificial mistakes. The percentage variation of SUV max (%∆ SUV max ) (1), SUVmean (%∆ SUV mean ) (2), and MTV (%∆ MTV) (3) were calculated. If the patient had more than one lesion, then only the highest SUV max at the activated bed was included. Furthermore, the lesions were divided into two groups according to their anatomical location. The upper lung group included bilateral upper lobes (S1-S3), and the lower lung group included bilateral middle and lower lobes (S4-S10).

Image Quality Assessment
A physician with over 10 years of experience in the nuclear medicine field subjectively reviewed and scored the image quality of the two reconstructions (QSC and NC). A 5-point Likert grading was applied to assess image sharpness and lesion distortion to judge the effects of artifacts. For image sharpness, the reader reviewed QSC and NC at the same time and rated the lesions as follows: 5 is an excellent image clearly showing lesions; 4 is a good image with minimal blurring lesion; 3 is an acceptable image; 2 is a poor image that affects diagnosis; 1 is a non-diagnostic image with severe blurring lesion. For lesion distortion, the reader compared the lesion in PET and CT images and assessed whether the PET lesion was distorted in terms of external morphology: 5 is excellent lesion external morphology and almost looks the same as the CT one; 4 is > 75% of lesion external morphology similar to CT; 3 is moderate distortion image with 50%-75% of lesion external morphology similar to CT; 2 is 25%-50% of lesion external morphology similar to CT; and 1 is severe distortion image with < 25% lesion of lesion external morphology similar to CT. The images were not scored when the CT image could not be diagnosed or the external morphology could not be defined, such as in atelectasis and tumor necrosis.

Statistical Analysis
VOI measured data were presented as mean ± standard deviation. Patients' characteristics and DDG Q.Static activation were assessed by Chi-square test. Differences of SUV max , SUV mean , and MTV for QSC and NC were assessed by paired sample t-test. Differences of SUV max , SUV mean , MTV, %∆ SUV max , %∆ SUV mean , and %∆ MTV between the two groups (upper and lower lung groups) were evaluated by Mann-Whitney U test. Correlation between %∆ SUV and MTV was evaluated by simple linear regression analysis. A P value of < 0.05 was considered to be statistically significant.

Results
Among the 194 reviewed patients, 124 (63.9%) had R values over the threshold and had activated the DDG Q.Static acquisition procedure in at least one bed. The majority of the activated beds were beds 4 (80/124) and 5 (91/124), which completely covered the lower lung and diaphragm areas. Meanwhile, bed 3(3/124) covering the upper lung was seldom activated (Table 1). No significant relationship was between the DDG Q.Static activation and the sex or BMI (P > 0.05). Among the 124 patients, only 49 had lesions located at the activated beds. The other 75 patients were excluded because they could not assess the correction ability of the DDG Q.Static. Only these 49 lesions were evaluated for quantitation accuracy and subjective image quality assessment. Table 2 shows the quantitative value of QSC and NC for quantitation accuracy. In most lesions, QSC had significantly higher SUV max and SUV mean and less MTV than NC. The mean values of SUV max , SUV mean , and MTV were 9.72 ± 5.42, 6.48 ± 3.68, and 5.55 ± 7.33 in QSC, respectively, and 9.20 ± 5.42, 6.11 ± 3.68, and 6.54 ± 8.58 in NC, respectively (P < 0.05). The %∆ SUV max , %∆ SUV mean , and %∆ MTV were 8.54% ± 11.62%, 8.65% ± 11.49%, and − 14.79% ± 20.70%, respectively. For subjective image quality assessment, four scores were obtained for each lesion after being reviewed by the reader. However, two cases were not scored for lesion distortion because external morphology cannot be defined in their CT images. The average scores for image sharpness and lesion distortion were 3.71 ± 0.61 and 3.14 ± 0.65 in QSC, respectively, and 3.08 ± 0.75 and 2.98 ± 0.76 in NC, respectively. Overall, the image quality of QSC scored higher than that of NC in image sharpness but not in lesion distortion. Compared with NC, QSC improved by 1.18 and 1.03 times in image sharpness and lesion distortion, respectively (Fig. 2).

Discussion
In this study, the quantitative values in QSC were significantly higher than those in NC as shown in Table 2 (P < 0.05). The image scores in QSC were also higher than those in NC (Fig. 2). Therefore, most of the lesions had significantly increased SUV (increasing 8.54% ± 11.62% in SUV max and 8.65% ± 11.49% in SUV mean ) and decreased   Fig. 3. A 68-year-old male with a right lower lobe lesion underwent PET/CT scan. NC presented blurred and unconcentrated lesion and distorted lesion boundary that expanded to the liver, thus causing diagnostic mistake and quantitative underestimation. By contrast, QSC clearly presented the lesion and was accurately concentrated on the lung. Furthermore, the quantitative values were improved: SUV max increased from 6.62 to 7.50 (13.29%), SUV mean increased from 4.12 to 4.84 (17.48%), and MTV was reduced from 4.23 cm 3 to 1.59 cm 3 (− 62.41%). Therefore, the DDG Q.Static technique could provide enhanced image quality and accurate quantitative values.
Among the patients in this study, 124 (63.9%) underwent DDG Q. Static procedure. The activated location included at least the fourth or fifth axial field of view (beds 4 and 5). This result could be interpreted as follows: respiratory motion seemed to be evident at bed 4 or 5 that almost covered the lower lung and diaphragm areas. For the quantitative value between the upper and lower lung groups, the results showed that the variation percentage of quantitative values in the lower lung group was significantly higher than those in the upper lung group. On the basis of the above two results, the lower lung and diaphragm areas might have been moved violently by free breathing, thus producing severe respiratory motion artifacts that degraded the PET accuracy. Many past studies have shown similar results [16,17]. Therefore, correcting these artifacts is important, especially when the lesions are located in the lower lung area.
Not all the patients presented similar results. A small number of patients showed decreased SUV or increased MTV after the DDG Q.Static correction. Two possible situations might have influenced the correction effect. First is that the lesions might have crossed two acquisition beds or are located at the overlapped area. When one of the acquisition beds activated the DDG Q.Static technique and the other did not, the result would indicate that the lesion was corrected partially. Therefore, the quantitative values might have been affected. Second is that the patients breathed irregularly and unstably. Since the DDG Q.Static has set fixed parameters with 30% phase offset and 50% phase width to acquire the stable end-expiration phase, the irregular breathing method might affect the calculation to extract the stable phase. Therefore, the end-expiration phase data might not be acquired with unstable breathing patterns, and the corrected result might be varied. A previous study used different values of phase offset and phase width to compare with the default ones. They demonstrated that the use of the variable phase offset and phase width that were optimized to each patient's individual breathing cycles was superior to the use of fixed phase offset and phase width [18]. Although the default parameters are not optimal for all patients, they are suitable for most people's breathing styles and thus can be applied to extract the stable end-expiration phase. If patients need individual parameters to fit their breathing styles, then the DDG Q.Static provides a retrospective modified option to extract the stable phase with suitable parameters suitable for the patients in need.
In the subjective image quality assessment, the lesions themselves were evaluated rather than the overall image quality because we focused on assessing whether the DDG Q.Static technique decreases the level of lesion blur. For image sharpness, the results showed that most of the lesions in QSC scored higher than those in NC because the DDG Q.Static decreased the effect of respiratory motion artifacts and made the images concentrated and clear. For lesion distortion, the results showed no significant difference between QSC and NC. Given that most of the lesions' edges were clearly presented in NC, no significant different was observed after the DDG Q.Static correction. However, some lesions were still located in the vigorous motion area (such as lower lobe or close to diaphragm), which usually presented blurred and distorted edges before correction. In this condition, the scores of QSC would be higher than those of NC. In summary, the subjective image quality scores confirmed that the DDG Q.Static technique can reduce blur and improve the lesion's conspicuity.

Conclusion
The DDG Q.Static technique is a convenient and powerful tool to eliminate/correct respiratory motion-associated blur. Without the need for external devices, this method can easily obtain respiratory information and calculate gated images from PET list data. Hence, technologists can shorten the time of positioning patients to reduce radiation exposure.
Our study showed that the DDG Q.Static technique can eliminate motion artifacts and improve the quantitative values (high SUV and lower metabolic tumor volume) and image quality to upgrade the PET accuracy. In particular, the lesions located at the lower lung or close to the diaphragm areas, which are recognized as violent motion areas, must be corrected. In summary, the DDG Q.Static is a valuable and clinically helpful technique in PET/CT scans.