Correcting CT misregistration in data-driven gated (DDG) PET with PET self-gating and deformable image registration

Abstract

Background:

Misregistration between CT and PET data can result in mis-localization and inaccurate quantification of functional uptake in whole body PET/CT imaging. This problem is exacerbated when an abnormal inspiration occurs during the free-breathing helical CT (FB CT) used for attenuation correction of PET data. In data-driven gated (DDG) PET, the data selected for reconstruction is typically derived from the end-expiration (EE) phase of the breathing cycle, making this potential issue worse.

Purpose:

The objective of this study is to develop a deformable image registration (DIR)-based respiratory motion model to improve the registration and quantification between misregistered FB CT and PET.

Methods:

Twenty-two whole-body ¹⁸F-FDG PET/CT scans encompassing 48 lesions in misregistered regions were analyzed in this study. End-inspiration (EI) and EE PET data were derived from −10 to 15% and 30 to 80% of the breathing cycle, respectively. DIR was used to estimate a motion model from the EE to EI phase of the PET data. The model was then used to generate PET images at any phase of up to four times the amplitude of motion between EE and EI for correlation with the misregistered FB CT. Once a matched phase of the FB CT was determined, FB CT was deformed to a pseudo CT at the EE phase (DIR CT). DIR CT was compared with the ground truth DDG CT for AC and localization of the DDG PET.

Results:

Between DDG PET/FB CT and DDG PET/DIR CT, a significant increase in $\text{Δ\%SUV}$ was observed (p<0.01), with median values elevating from 26.7% to 42.4%. This new method was most effective for lesions ≤3 cm proximal to the diaphragm (p<0.001) but showed decreasing efficacy as the distance increased. When FB CT was severely misregistered with DDG PET (>3cm), DDG PET/DIR CT outperformed DDG PET/FB CT alone (p<0.05). Even when patients showed varied breathing patterns during the PET/CT scan, DDG PET/DIR CT still surpassed the efficiency of DDG PET/FB CT (p<0.01). Though DDG PET/DIR CT couldn’t match the performance of the DDG PET/CT ground truth (42.4% vs. 53.6%, p<0.01), it reached 84% of its quantification, demonstrating good agreement and a strong overall correlation (regression coefficient of 0.94, p<0.0001). In some cases, anatomical distortion and blurring, and misregistration error were observed in DIR CT, rendering it still unable to correct inaccurate localization near the boundaries of two organs.

Conclusions:

Based on the motion model derived from gated PET data, DIR CT can significantly improve the quantification and localization of DDG PET. This approach can achieve a performance level of about 84% of the ground truth established by DDG PET/CT. These results show that self-gated PET and DIR CT may offer an alternative clinical solution to DDG PET and FB CT for quantification without the need for additional cine-CT imaging. DIR CT was at times inferior to DDG CT due to some distortion and blurring of anatomy and misregistration error.

INTRODUCTION

Misregistration and respiratory motion are well-known problems in positron emission tomography/computed tomography (PET/CT) imaging. In the past, the only option for respiratory gating in PET and CT was through external device-based gating (EDG) [1]. This method involves proper selection and screening of patients, setting up hardware, and making alterations to the usual clinical procedure.

A solution known as data-driven gated (DDG) PET is now commonly used in clinical settings as an alternative to EDG methods. This solution directly obtains respiratory motion information from raw PET data without the need for an external device [2, 3]. According to previous research, motion correction through DDG can lead to an improvement in standard uptake values (SUV) by up to 25% and produce more precise representations of uptake [4]. The CT scan in PET/CT is typically a helical acquisition obtained under free-breathing (FB) conditions. As a result, it should be noted that even when using DDG for PET scans, there is still a risk of misregistration between DDG PET and the FB CT data used for attenuation correction (AC) and localization. This issue has been previously acknowledged in studies on DDG PET, as misregistration between DDG PET and helical CT can occur in certain cases [2, 5]. The use of incorrect CT phases for AC can lead to significant changes in PET quantification, with the impact more severe for DDG PET than static PET [6].

We recently implemented a new method that combines DDG CT with DDG PET to overcome the problems of PET/CT misregistration and inaccurate PET quantification [7]. This new DDG PET/CT involves a cine-CT in the region of misregistration and has been implemented on a network of seven GE PET/CT scanners at our institution. From the cine-CT, DDG CT was extracted to provide improved registration with DDG PET. This strategy offers a lower CT radiation dose and a shorter scan time than repeat PET/CT in cases of clinically relevant misregistration [8]. The resulting DDG PET/CT also led to improved registration and enhanced PET quantification through mostly increased SUV. These results showcase the potential of DDG PET/CT for significant clinical impact when both misregistration and motion correction are accounted for together in PET/CT [8].

Respiratory motion models based on image registration algorithms have been previously proposed and applied to medical images for motion correction [9]. Specifically, there are research efforts to register PET and CT using deformable image registration (DIR) [10]. However, direct, non-rigid image registration between CT images with high resolution and low noise, and PET images with low resolution and high noise, remains challenging. A recent alternative approach proposed was to register the CT to a reference phase selected from multiple DDG PET phases [11, 12]. However, because the selected reference phase was one of the DDG PET phases between EE and EI phases, it may only work well when the CT phase was also within these two phases. In addition, the quality of the registration was greatly limited by PET image noise from low counts in the multi-phase gating process.

We recently offered an approach to register CT to DDG PET using a linear motion model derived from DDG PET phases [13]. In this study, our aims were to (1) develop a DIR-based respiratory motion model derived from self-gated PET data (DIR CT), and (2) compare DDG PET in quantification and registration using DIR CT and ground-truth DDG CT for AC.

METHODS

Patient Selection

The study was approved by an Institutional Review Board. The need for informed written consent was waived for this study. Twenty-two whole-body 18F-fluorodeoxyglucose (FDG) PET/CT scans that included a cine-CT scan [8] for misregistration correction were analyzed in this study.

PET/CT Acquisition Protocols

The scans were acquired from four GE Discovery MI scanners with a 25-cm detector along the table direction. The injected activity was targeted at 370 MBq of ¹⁸F-FDG. The FB CT scan was acquired first with the following parameters: 120 kVp, 60–560 mA tube current, tube current modulation (TCM), noise index=30, 1.375 pitch, 0.5 s gantry rotation time, and a 64×0.625 mm (4 cm) beam width. Whole-body PET was acquired with 2, 2.5 and 3 min per bed position for body mass index (BMI) < 35, 35–40 and >40, respectively.

Upon identification of clinically relevant PET/CT misregistration by the imaging technologist, a cine-CT was acquired over the misregistered region after the PET acquisition for misregistration correction. The cine-CT scan protocol was 120 kVp, 5 s cine scan duration, gantry rotation time = 0.8 s, X-ray collimation = 8 × 2.5 mm (2 cm), noise index = 70, minimum mA = 10, and maximum mA = 20. The estimated effective radiation dose from the cine-CT was 1.3 ± 0.6 mSv [7].

DDG PET and DDG CT

The listmode PET data were retrospectively processed for DDG-PET using Duetto (GE Healthcare), a PET toolbox equipped with MATLAB and MATLAB-callable C/C++ functions for the task of unlisting, correcting, reconstructing and analyzing PET listmode and raw sinogram data. The methodology of DDG PET derivation was outlined previously [7]. The respiratory cycle was defined as the interval between two consecutive end-of-inspiration points. Within each cycle, phases with minimal motion are identified based on motion amplitude. DDG PET of the EE phase was derived from 30% to 80% of this respiratory cycle, corresponding to a nearly stationary period around the end of expiration. DDG PET of the EI phase was derived from the last 10% of the preceding cycle to the first 15% of this cycle (−10%−15%), corresponding to a period with minimal motion around the end of inspiration. All retrospective PET images were reconstructed with time-of-flight ordered subsets expectation maximization, including point-spread function modeling (VPFX-S). Reconstructions utilized 17 subsets and 2 iterations, with a 5 mm Gaussian post-filter.

The cine-CT data was used to derive DDG CT at the EE phase, which replaced the same region in the FB CT for improved AC of the DDG PET data. The details related to EE CT derivation, as well as PET, FB CT, and cine-CT acquisitions, can be found in our previous work [7, 8].

DIR CT Workflow

Figure 1 provides a schematic workflow of the methodology used to derive DIR CT from self-gated PET data.

Figure 1.

Flowchart of the proposed method for improving registration between FB CT and data-driven gated (DDG) PET. DVF: deformation vector field; EE: End Expiration; EI: End Inspiration; DIR: Deformable Image Registration; FB: Free Breathing.

To establish the motion model, the PET list-mode data underwent a principal components analysis (PCA)-based DDG process to extract EI and EE data from the −10% to 15% and 30% to 80% phases of the respiratory cycle, respectively. To estimate a deformation vector field (DVF) between the EE and EI phases of the PET data, a multi-resolution deformable registration tool from Duetto (GE Healthcare) was employed [14]. Subsequently, the DVF produced by the DIR tool underwent linear interpolation to generate DVFs at intermediate phases between EE and EI. The DVFs were also extrapolated up to four times the amplitude from EE to EI to create additional phases outside the EE and EI limits. A range of PET images at other phases were generated via application of the interpolated or extrapolated DVFs to the EE phase PET image.

An 8-cm line, manually centered at the midpoint of the right diaphragm, was employed to create line profiles from both misregistered FB CT and the collection of PET images at different phases created by the DVFs. The line profile extracted from misregistered FB CT was correlated with the line profiles from the PET images and the correlation coefficients were used to generate a correlation curve. The matched phase of the PET images with the FB CT was identified as the phase with the highest correlation coefficient. The DVF of the matched phase was then inverted to transform the FB CT to an EE CT – identified hereafter as DIR CT. This newly generated DIR CT was compared with FB CT and DDG CT for AC of DDG PET.

Quantitative Measurements

Four different PET/CT methods were compared: (1) BL PET/FB CT (baseline PET, AC by FB CT), (2) DDG PET/FB CT (DDG PET, AC by FB CT), (3) DDG PET/CT (DDG PET, AC by DDG CT), and (4) DDG PET/DIR CT (DDG PET, AC by DIR CT). The PET data, CT used for AC and their associate respiratory phases are outlined in Table 1.

Table 1.

Summary of PET/CT methods.

PET/CT method	PET Data	PET Phase	CT for Attenuation Correction	CT Phase
BL PET/FB CT	Baseline PET	Average	FB CT	Random
DDG PET/FB CT	DDG PET	End-Expiration	FB CT	Random
DDG PET/CT	DDG PET	End-Expiration	DDG CT	End-Expiration
DDG PET/DIR CT	DDG PET	End-Expiration	DIR CT	End-Expiration

BL: baseline; FB: free-breathing; DDG: Data-Driven Gated, DIR: Deformation Image Registration.

Only lesions located in the range of the cine-CT scan were compared. In total, 48 clinically identified lesions were found in the misregistered regions, comprising 18 liver, 26 lung, 3 lymph node and 1 esophagus lesions.

Maximum SUV (SUV_max) was computed for all lesions. The percentage change ($\text{Δ\%SUV}$) as defined in Equation 1, in comparison to BL PET/FB CT was calculated for all other PET/CT datasets.

$$\Delta \mathrm{\%}SUV=\frac{({SUV}_{max}-SU{V}_{max}^{BLPET/FBCT})}{SU{V}_{max}^{BLPET/FBCT}}\times 100\mathrm{\%}$$ (1)

$\text{Δ\%SUV}$ was quantified for all lesions across all methods. The efficacy of each method was compared with DDG PET/CT, which served as the ground truth. Additional variables, such as FB CT phase, severity of misregistration between FB CT and DDG PET and distance of lesions to the diaphragm at the EE phase, were also investigated as detailed in the subsequent paragraphs.

In this study, most patients showed a deeper inspiration during the CT scan than average breathing during the follow-up PET scan. Consequently, the matched phase of FB CT may align with an extrapolated PET phase, which was outside the range of the EE and EI phases. To assess the impact of breathing irregularity on the $\text{Δ\%SUV}$ of all methods, comparisons were made when the FB CT matched best with an extrapolated PET phase.

To determine the impact of misregistration between FB CT and DDG PET at EE, all methods were evaluated for severity of misregistration. A fusion image of DDG PET at EE and the matched FB CT phases was generated in MIM (MIM Software Inc.). The edge detection of MIM was employed to delineate the liver/lung boundary, and the misregistration from EE to the matched FB CT phase was measured between these boundaries at the center of the right diaphragm. Misregistration less than 3cm were considered ‘mild,’ while those greater than 3cm were identified as ‘severe.’ All the methods were compared with respect to this categorization of misregistration.

The $\text{Δ\%SUV}$ of the methods was also assessed on the basis of lesion distance from the diaphragm, as determined from the lesion’s center to the diaphragm on fused DDG PET/CT. The lesions were grouped into three categories: proximal (< 3 cm), intermediate (3 – 6 cm), and distal (> 6 cm). Additionally, a linear regression was conducted to discern the relationship between model $\text{Δ\%SUV}$ and lesion distance to the diaphragm.

Statistical Analysis

The majority of the datasets analyzed were right-skewed and did not pass Shapiro-Wilk normality tests, thus comparisons were made using non-parametric significance tests in most instances. For all matched group comparisons or repeated measures, Friedman’s test was employed with a false discovery rate correction using the two-state step-up method of Benjamini et al. [15]. Multiple comparisons were assessed for more specific analysis among groups. Pearson’s correlation coefficient and Bland-Altman style plots were used for evaluating the correlation and agreement with the ground truth. All statistical analyses were performed with GraphPad Prism 9.0.0 (GraphPad Software), and statistical significance was considered true for p < 0.05.

RESULTS

Table 2 provides summary statistics for all the methods and group comparisons. Values in this table are expressed as medians, with the interquartile range (IQR) defined as the difference between the third quartile (Q3) and the first quartile (Q1).

Table 2.

Statistical summary of percentage change ($\text{Δ\%SUV}$) for all methods

	A: DDG PET/FB CT	B: DDG PET/DIR CT	C: DDG PET/CT	p value
				A vs. B	A vs. C	B vs. C
Total (n=48; c=22)	26.7 (8.7–44.1)	42.4 (21.6–72.9)	53.6 (24.7–81.3)	**	****	**
Lesion Distance
Proximal (0–3 cm, n=20)	30.8 (8.7–42.2)	66.2 (36.9–103.3)	73.4 (46.4–110.6)	***	****	*
Intermediate (3–6 cm, n=18)	23.5 (6.3–57.2)	33.5 (16.5–64.6)	50.5 (17.9–71.9)	*	****	**
Distal (>6 cm, n=10)	30.0 (17.0–37.7)	26.7 (12.5–43.9)	28.6 (15.5–54.6)	ns	ns	ns
Breath Irregularity (n=40; c=17)	22.6 (5.9–34.32)	34.3 (18.6–72.9)	53.0 (21.2–84.4)	*	****	***
Misalignment
Mild (≤3 cm, n=28, c=14)	25.6 (8.7–48.1)	36.6 (17.1–63.1)	46.1 (21.2–68.5)	ns	****	ns
Severe (>3 cm, n=20, c=8)	30.8 (7.3–39.3)	53.6 (26.0–112.6)	68.8 (42.4–142.4)	*	****	**

All values are $\text{Δ\%SUV}$, calculated relative to BL PET/FB CT

All values are displayed as median (Q1-Q3)

n: number of lesions; c: number of cases

Lesion Distance: Proximal (≤ 3 cm), Intermediate (3–6 cm) and Distal range: (≥6 cm)

Breathing Irregularity: Extrapolated Phase of FB CT (beyond EI)

Misalignment from FB CT to PET EE: Mild (0–3 cm) and Severe (3–6 cm)

^*p < 0.05;

^**p < 0.01;

^***p < 0.001;

^****p < 0.0001; ns, not significant

Comparison of DDG PET/FB CT, DDG PET/DIR CT and DDG PET/CT

All 48 lesions from 22 patients were included in this comparison. Median percentage change for DDG PET/FB CT, DDG PET/DIR CT and DDG PET/CT relative to BL PET/FB CT were 26.7% (IQR 8.7%−44.1%), 42.4% (IQR 21.6%−72.9%) and 53.6% (IQR 24.7%−81.3%). Statistically significant differences were noted in $\text{Δ\%SUV}$ between DDG PET/FB CT and DDG PET/DIR CT (p<0.01), between DDG PET/FB CT and DDG PET/DIR CT (p<0.0001), and also between DDG PET/DIR CT and DDG PET/CT (p<0.01; Figure 2A).

Figure 2.

Comparative analysis of lesion $\text{Δ\%SUV}$ across all approaches and the agreements with ground truth. (A) Comparison of $\text{Δ\%SUV}$ of all lesions between DDG PET/FB CT, DDG PET/DIR CT, and DDG PET/CT. (B) Correlations between DIR CT and DDG CT and between FB CT and DDG CT ($\text{Δ\%SUV}$). (C) Bland-Altman plot showed weak agreement between DDG PET/FB CT and DDG PET/CT and (D) Strong agreement between DDG PET/DIR CT and DDG PET/CT. For readability in this figure: ‘FB CT’ denotes ‘DDG PET/FB CT’; ‘DIR CT’ denotes ‘DDG PET/DIR CT’; and ‘DDG CT’ denotes ‘DDG PET/DDG CT’. In the boxplot, central line indicates the median; box edges show Q1 and Q3 quartiles. Whiskers extend to the minimum and maximum data values. *: p<0.05; ****: p<0.0001

The correlation coefficient of $\text{Δ\%SUV}$ between DDG PET/DIR CT and DDG PET/CT was higher than that between DDG PET/FB CT and DDG PET/CT (r = 0.94, p < 0.0001 vs. r = 0.41, p = 0.004; Figure 2B). The increase in quantification relative to BL PET/FB CT from DDG PET/DIR CT was ~84% of that provided by DDG PET/CT, on average (slope=0.84).

In the Bland-Altman plot, the agreement of $\text{Δ\%SUV}$ between DDG PET/DIR CT and DDG PET/CT (Figure 2D) was also higher than that between DDG PET/FB CT and DDG PET/CT (Figure 2C).

Overall, motion correction with DDG PET/FB CT resulted in a median increase in SUV_max of 26.7% relative to BL PET/FB CT. Then in DDG PET/DIR CT and DDG PET/CT, misregistration was also corrected along with lesion motion because the CT used for AC better matched the phase of PET. As a result, significant increases in SUV_max were observed relative to DDG PET/FB CT. For most lesions, SUV_max increased continuously from BL PET/FB CT → DDG PET/FB CT → DDG PET/DIR CT → DDG PET/CT. In one patient study, SUV_max values of a subdiaphragmatic liver lesion demonstrated this trend, as shown in Figure 3. However, for another lateral liver lesion close to the chest wall where both motion and misregistration are restricted, DIR CT failed to further improve PET quantification.

Figure 3.

Visualization of PET/CT registration and SUV quantification across different methods. (A) BL PET/FB CT: notable misregistration and lowest SUV_max. (B) DDG PET/FB CT: Marginal SUV_max improvement, though misregistration persists. (C) DDG PET/DIR CT: improved registration and SUV quantification. (D) DDG PET/CT: the ground truth with improved registration and SUV quantification. Note: For each method, the corresponding SUV_max value is next to the lesion.

Figure 4 demonstrated the capabilities and limitations of DIR CT in lesion localization through three example cases. Figure 4A illustrated the most representative outcome where DIR CT showed improvement over FB CT in terms of lesion localization, with comparable performance to that of DDG CT. Figure 4B represented one scenario where registration inaccuracy was amplified when excessive extrapolation was needed to match FB CT. This was observed in 7 cases where the matched phase exceeded 2.5 times of the PET EE to EI range. In this example case, DIR CT registration error and regional anatomical distortions resulting from the amplified extrapolation produced a slightly overcorrected misregistration (green arrows). Despite this, DIR CT still improved SUV quantification over FB CT. Figure 4C introduced one challenging case, with both the conventional FB CT and the enhanced DIR CT unable to precisely localize a lesion within the gallbladder. Together, these cases highlight both the strengths and areas of potential refinement for the DIR CT methodology in PET/CT imaging.

Figure 4.

Illustration of DIR CT’s improvement and limitations in lesion localization using three cases. (A) A definite improvement in lesion localization using DIR CT over FB CT, comparable with DDG CT. (B) A slight overcorrection observed with DIR CT, resulting in residual misregistration yet retains an advantage over FB CT. (C) A situation where both FB CT and DIR CT fail to accurately localize a lesion in the gallbladder. Regional anatomical distortions were highlighted by green arrows. For readability in this figure: column name ‘FB CT’ denotes ‘DDG PET/FB CT’; column name ‘DIR CT’ denotes ‘DDG PET/DIR CT’; and column name ‘DDG CT’ denotes ‘DDG PET/DDG CT’.

Breathing Irregularity

Of the 22 patients, 17 required an extrapolated phase outside of the EE to EI range to match with FB CT. Figure 5 shows the results of quantification changes when an extrapolated phase was required to best match the FB CT. Significant differences in quantification between the various methods were observed when the diaphragm position in FB CT was identified as outside the EI phase.

Figure 5.

Evaluation of $\text{Δ\%SUV}$ when FB CT was at an extrapolated phase, which was a phase beyond EI. No significant change was observed when FB CT was an interpolated phase. Both DDG PET/DIR CT and DDG PET/CT showed $\text{Δ\%SUV}$ increase. For readability in this figure: column name ‘FB CT’ denotes ‘DDG PET/FB CT’; column name ‘DIR CT’ denotes ‘DDG PET/DIR CT’; and column name ‘DDG CT’ denotes ‘DDG PET/DDG CT’. In the boxplots, central line indicates the median; box edges show Q1 and Q3 quartiles. Whiskers extend to the minimum and maximum data values. n: number of lesions; c: number of cases. *: p<0.05; **: p<0.01; ***: p<0.001; ****: p<0.0001

Severity of FB CT Misregistration

Figure 6 shows results for changes in PET quantification on the basis of misregistration severity. As seen in Figure 6A, DDG PET/DIR CT showed a modest, non-significant increase in $\text{Δ\%SUV}$ over DDG PET/FB CT (36.6% vs. 25.6%, p>0.05) for cases with mild misregistration. In comparison, DDG PET/CT exhibited a statistically significant increase (46.1% vs. 25.6%, p<0.001). On the other hand, for severe misregistration cases in Figure 6B, both DDG PET/DIR CT and DDG PET/CT maintained significant SUV_max increases over DDG PET/FB CT alone.

Figure 6.

$\text{Δ\%SUV}$ comparison of three approaches when FB CT was at a mild (≤3cm) or severe (>3cm) misregistration from PET EE. Misregistration was defined as the distance that the diaphragm moved from EE to the matched phase of FB CT. DDG PET/CT show significant improvement with mild misregistration (A), while both DDG PET/DIR CT and DDG PET/CT showed significant improvements with severe misregistration (B). For readability in this figure: column name ‘FB CT’ denotes ‘DDG PET/FB CT’; column name ‘DIR CT’ denotes ‘DDG PET/DIR CT’; and column name ‘DDG CT’ denotes ‘DDG PET/DDG CT’. In the boxplot, central line indicates the median; box edges show Q1 and Q3 quartiles. Whiskers extend to the minimum and maximum data values. n: number of lesions; c: number of cases. *: p<0.05; **: p<0.01; ***: p<0.001; ****: p<0.0001

Lesion Distance to Diaphragm

Figure 7 shows results for changes in PET quantification with lesion distance to the diaphragm. For lesions near the diaphragm (proximal; ≤3 cm), both DDG PET/DIR CT and DDG PET/CT increased SUV_max significantly more than DDG PET/FB CT (p<0.001 and p<0.0001). Lesions at an intermediate distance (3–6 cm) only maintained a significant difference to DDG PET/FB CT with DDG PET/CT (p<0.0001). DDG PET/CT increased SUV_max significantly more than DDG PET/DIR CT in both the proximal and intermediate groups. For distal lesions the furthest away from the diaphragm (≥6 cm), there were no significant differences among the three methods.

Figure 7.

The three approaches were compared at various lesion locations. The lesion location was measured from the center of the lesion to the diaphragm and labeled as Proximal (≤3cm), Intermediate (3–6 cm), and Distal (≥ 6cm). For readability in this figure: column name ‘FB CT’ denotes ‘DDG PET/FB CT’; column name ‘DIR CT’ denotes ‘DDG PET/DIR CT’; and column name ‘DDG CT’ denotes ‘DDG PET/DDG CT’. (A) DDG PET/DIR CT and DDG PET/CT worked more effectively at the proximal (A) and intermediate range (B), but no improvements were observed in the distal range (C). For DDG PET/DIR CT, improvements were linearly decreased with the distance of the lesion from the diaphragm (D). In the boxplots, central line indicates the median; box edges show Q1 and Q3 quartiles. Whiskers extend to the minimum and maximum data values. n: number of lesions. *: p<0.05; **: p<0.01; ***: p<0.001; ****: p<0.0001

DISCUSSION

In this work, we presented results supporting the idea that a motion model derived from self-gated PET can be applied to the traditional FB CT for improved AC in DDG PET applications. The DIR CT outlined in this work results from the deformable registration of FB CT to the EE phase for improved registration with DDG PET. The best-matched PET phase that aligns with the FB CT is selected from a range of available phases extending beyond the EE to EI range. This PET and FB CT phase matching serves as the foundation of the correction process, enabling DIR CT to mitigate misregistration between PET and FB CT and improve DDG PET quantification even in cases with severe misregistration. DDG PET/DIR CT showed a high degree of agreement with the ground-truth DDG PET/CT, achieving 84% of the level of SUV_max increase from DDG PET/CT, on average. This approach is particularly beneficial for lesions near the diaphragm where motion is more pronounced.

Previous work on DDG PET/FB CT reported a 10%−20% increase in SUV_max [2, 5, 16]. Our own findings suggested there was a 26.7% increase (Figure 2A and B), which is consistent with these prior studies. In the cases analyzed here, both DDG PET/CT and DDG PET/DIR CT showed significant improvement over DDG PET/FB CT (53.6% and 42.4%) due to improved registration and AC. To our knowledge, the increases in SUV_max observed here for DDG-PET/DIR CT were larger than any other study of BL PET/FB CT or DDG PET/FB CT [17]. The improvements in both localization and quantification indicate that DDG PET, whose data are captured at the EE phase, will benefit from a CT matched to the same phase. In terms of lesion distance to the diaphragm, our analysis showed more improvement for lesions in the proximal range (≤3 cm). These lesions in the proximal range suffered the most from misregistration, leading to the largest SUV_max increases with DIR CT at nearly twice the increases from DDG PET/FB CT alone.

Misregistration severity between DDG PET and FB CT also played a pivotal role in the PET quantification results. In Figure 6B for cases with severe misregistration, both DDG PET/DIR CT and DDG PET/CT showed significant improvement over DDG PET/FB CT alone. Employing DIR CT and DDG CT can lead to improved quantification, as both are more likely aligned with DDG PET at the EE phase. For lesions with mild misregistration, DDG PET/CT showed significant changes in SUV_max compared to DDG PET/FB CT, while the increase observed with DDG PET/DIR CT was not statistically significant (Figure 6A). As illustrated by the lateral liver lesion in Figure 3, lesions positioned closely aligned with the spinal cord or chest wall may experience less respiratory motion and mild misregistration. The potential advantages of registration from the DIR process might be offset by inaccuracies introduced by the motion model. In such cases, DIR CT might not impact PET quantification.

The inability of patients to breath consistently during PET/CT scans is an issue, yet it has not been extensively addressed in previous studies. In 17 of the 22 cases analyzed here, we observed patients beginning with deeper respirations during the initial CT scan, which then became shallower by the time the PET was acquired. This led to the FB CT phase falling outside the defined range of the PET EE and EI phases. The results in Figure 5 showed that both DIR CT and DDG CT were still able to significantly improve PET quantification for lesions in these 17 cases. The improved registration between CT and DDG PET as provided by both DIR CT and DDG CT can be beneficial in cases where consistent breathing control poses a challenge for certain patients.

Historically, misregistration between PET and FB CT has been handled in one of two ways: 1) accept the misregistration and provide no corrections, or 2) repeat both PET and CT over the misregistered region, hoping the second acquisition will be registered better [7]. Both DIR CT and DDG CT offer alternative solutions that can address misregistration between PET and FB CT and provide corrections to PET quantification. Neither requires a full repeat of both PET and CT, which will simplify clinical workflows and increase patient throughput. While the results in this study showed that DIR CT was not always as effective as DDG CT, its merits lie in its post-processing nature – DIR CT eliminates the additional radiation exposure of the patient from the cine-CT required for DDG CT. More importantly, DIR CT also requires no change to existing clinical workflows, making it a seamless addition to any clinic regardless of personnel training or staffing limitations. Nevertheless, the computational demand for generating DIR CT is notably higher compared to DDG CT. On average, creating DIR CT takes about four hours on a Linux server equipped with 256 GB RAM and an Intel Xeon Gold 5217 CPU. This extended processing time is primarily due to the program being developed in MATLAB and the intensive computing required for reconstructing whole-body DDG PET at the EE and EI phases. In contrast, DDG CT can be generated in 2 to 3 minutes on a standard desktop computer (Dell OptiPlex 7050, 8 GB RAM, i5–6500 CPU) [8]. This may present some limitations for extensive clinical utilization of DIR CT without enhanced computational resources.

This study has limitations that should be addressed. First, as the aim of this paper was to demonstrate the capability of our proposed method to correct PET/CT misregistration, only cases with compromised image quality due to misregistration were selected. The current DDG PET/CT clinical workflow at our institution is only initiated, and data only collected, when clinically relevant misregistration is identified during the initial PET/CT acquisition. To what extent the results presented here may apply to a more complete group of routine PET/CT scans is difficult to determine. However, our previous work did confirm that corrections from DDG CT remained relevant even in a cohort of patients and PET/CT scans that were collected without any knowledge of misregistration occurring a priori [18]. The second limitation is the potential registration error in the DIR process resulting from the inherently lower resolution, decreased contrast and higher noise in PET images compared to CT. In cases requiring excessive extrapolation (>2.5x the PET EE to EI range) to match the phase of FB CT, the imperfections in the DIR process will be amplified. This led to noticeable localized distortion on DIR CT images, as demonstrated in Figure 4B. Additionally, the transformation applied in the DIR process resulted in a smoother appearance of the DIR CT images, in contrast to the sharper FB CT and DDG CT images. While no loss of sensitivity for lesion localization was observed in this study, the blurring effect could potentially limit the visibility of fine structures and the detectability of very small lesions on DIR CT images. Although the PET quantification still largely benefits from this approach, the resulting DIR CT might not be appropriate for anatomical evaluations. In the end, any algorithm or correction process will likely be limited by the physical situation at hand: if patient breathing conditions are such that the anatomy during the FB CT is drastically different from that during the PET acquisition, optimized corrections are likely not attainable.

CONCLUSIONS

In this study, a new automatic DIR CT method was developed to address issues of misregistration in PET/CT imaging. DIR CT was based on a motion model derived from the self-gated PET data. This new approach could significantly improve the localization and quantification of DDG PET. When compared to DDG PET/FB CT alone, DDG PET/DIR CT was particularly effective for lesions proximal to the diaphragm where motion is most pronounced. This approach also showed consistent improvements in cases with severe misregistration between PET and CT, as well as cases with patient breathing inconsistencies between CT and PET acquisitions. The new DIR CT can be seamlessly integrated into existing clinical workflows. Without any modification or additional acquisition, on average it can provide 84% of the same improvements to PET quantification as the ground-truth DDG PET/CT that utilizes DDG CT extracted from a low dose cine-CT.