Impact of 18F-FDG PET/MR based tumor delineation in radiotherapy planning for cholangiocarcinoma

Abstract

Purpose

Radiation therapy (RT) is an effective treatment for unresectable cholangiocarcinoma (CC). Accurate tumor volume delineation is critical in achieving high rates of local control while minimizing treatment-related toxicity. This study compares ¹⁸F-FDG PET/MR to MR and CT for target volume delineation for RT planning.

Methods

We retrospectively included 22 patients with newly diagnosed unresectable primary CC who underwent ¹⁸F-FDG PET/MR for initial staging. Gross tumor volume (GTV) of the primary mass (GTV_M) and lymph nodes (GTV_LN) were contoured on CT images, MR images, and PET/MR fused images and compared among modalities. The dice similarity coefficient (DSC) was calculated to assess spatial coverage between different modalities.

Results

${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ (median: 94 ml, range 16–655 ml) was significantly greater than ${\text{GTV}}_{\text{M}}^{\text{MR}}$ (69 ml, 11–635 ml) (p = 0.0001) and ${\text{GTV}}_{\text{M}}^{\text{CT}}$ (96 ml, 4–564 ml) (p = 0.035). There was no significant difference between ${\text{GTV}}_{\text{M}}^{\text{CT}}$ and ${\text{GTV}}_{\text{M}}^{\text{MR}}$ (p = 0.078). Subgroup analysis of intrahepatic and extrahepatic tumors showed that the median ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was significantly greater than ${\text{GTV}}_{\text{M}}^{\text{MR}}$ in both groups (117.5 ml, 22–655 ml vs. 102.5 ml, 22–635 ml, p = 0.004 and 37 ml, 16–303 ml vs. 34 ml, 11–207 ml, p = 0.042, respectively). The ${\text{GTV}}_{\text{LN}}^{\text{PET}/\text{MR}}$ (8.5 ml, 1–27 ml) was significantly higher than ${\text{GTV}}_{\text{LN}}^{\text{CT}}$ (5 ml, 4–16 ml) (p = 0.026). GTV^PET/MR had the highest similarity to the GTV^MR, i.e., DSC^PET/MR-MR (0.82, 0.25–1.00), compared to DSC ^PET/MR-CT of 0.58 (0.22–0.87) and DSC^MR-CT of 0.58 (0.03–0.83).

Conclusion

¹⁸F-FDG PET/MR-based CC delineation yields greater GTVs and detected a higher number of positive lymph nodes compared to CT or MR, potentially improving RT planning by reducing the risk of geographic misses.

Introduction

Cholangiocarcinoma (CC) is the most common biliary tract and the second most common hepatobiliary malignancy, with a globally increasing incidence [1]. CC is often aggressive with poor overall prognosis due to diagnosis at an advanced-stage at presentation [2].

Treatment of CC is tailored based on the clinical stage at presentation. Complete surgical resection is considered the only curative intent therapy option, with a disease-free survival ranging from 12 to 36 months. The success of surgical resection varies based on multiple factors, including tumor location and characteristics such as size and margin, lymph node involvement, and postoperative complications [3, 4]. Only a small portion of patients present at an early stage of the disease that would make them good candidates for surgical resection [5]. In the vast majority of patients, tumors are locally advanced and unresectable at presentation; for these patients, treatment options include a combination of systemic chemotherapy and local therapies such as external beam radiation therapy (RT) [6, 7].

Of the local therapies, RT is the most extensively studied and has been shown to improve patient outcomes [8, 9]. A higher prescribed radiation dose is associated with improved patient outcomes [9–12], such as overall survival (OS) and local control [11, 13]. Furthermore, neoadjuvant chemoradiation regimens with higher radiation doses increases the chance of curative resection in previously inoperable patients, with survival outcomes comparable to those with resectable tumors at baseline [14]. To maximize tumor control yet minimize treatment-related toxicity, an accurate determination of tumor location and extension is essential to treat and improve the safety and effectiveness of RT [15, 16].

Metabolic imaging using ¹⁸F-fluorodeoxyglucose Positron Emission Tomography (¹⁸F-FDG PET) is a useful diagnostic tool for initial staging of CC, as these tumors often show high ¹⁸F-FDG avidity [17–20]. PET/CT-based RT planning has been shown to improve target volume delineation in several types of malignancies [21–23], including CC [24], by reducing geographic misses and potentially sparing the organs at risk (OAR).

Given the key role of RT in the management of CC, especially the importance of the radiation dose to the tumor in patient outcomes, improving RT planning by employing more accurate techniques is of great importance. PET/MR imaging is a hybrid imaging which enables simultaneous acquisition of the tumor metabolic data using PET and detailed anatomical data with exquisite soft tissue contrast using MR. Given that the metabolic and anatomic data are complimentary, PET/MR can markedly improve lesion delineation and potentially improve the RT planning [20]. A number of recent studies have suggested that PET/MR is more effective than PET/CT in RT planning in various cancers [25–27]. Therefore, ¹⁸F-FDG PET/MR could potentially improve RT planning over currently used imaging techniques such as CT, MR or ¹⁸F-FDG PET/CT. However, to our knowledge, the use of PET/MR has not specifically been assessed for RT planning in CC and this study is the first to assess PET/MR in RT planning of CC. In this study we compare ¹⁸F-FDG PET/MR-based tumor and lymph node delineation to CT- and MR-based techniques for RT planning in CC patients prior to treatment.

Methods

Patient population

The institutional review board approved the study protocol, and a waiver of informed consent requirement was obtained given the study’s retrospective nature. All steps of the study were compliant with HIPAA. We retrospectively reviewed all CC patients at Massachusetts General Hospital who underwent ¹⁸F-FDG PET/CT imaging from August 2017 to June 2019. We identified all the subjects with primary CC at presentation, who underwent ¹⁸F-FDG PET/MR for initial staging at our hepatic oncology referral center. None of the patients had undergone any previous treatment.

PET/MR and CT imaging protocol

All patients fasted for at least 6 hours prior to ¹⁸F-FDG injection. Their blood glucose level was confirmed to be less than 140 mg/dL prior to intravenous injection of 555–925 megabecquerel of ¹⁸F-FDG. When PET/CT and PET/MR was performed on the same day, patients first underwent ¹⁸F-FDG PET/CT imaging (Siemens Biograph PET/CT scanner, Siemens Medical Solutions, Erlangen, Germany). Approximately 60 min after PET/CT imaging, whole-body PET/MR was performed, followed by a focused upper abdominal protocol, using a whole-body PET/MR scanner (mMR, Siemens Healthcare, Erlangen, Germany). For the upper abdominal MR imaging, patients stayed in supine position with the arms by their sides. 0.1 mmol/kg (0.5 mmol/ml) of Gadoterate meglumine (Gd-DOTA - Dotarem, Guerbet, Princeton, NJ, USA) or Gadopentate dimeglumine (Gd-DTPA - Magnevist, Bayer Pharma AG, Berlin, Germany) was injected (3 ml/s followed by same volume saline at 3 ml/s) to assess the dynamic liver enhancement. PET scan was performed with a 4-min acquisition per bed position and four to six bed positions depending on patients’ height. PET images were reconstructed in a 258*258 matrix, using an AW OSEM 3D iterative algorithm in 3 iterations, and 21 subsets (voxel size 2.0*2.0*2.0 mm). Attenuation correction was automatically computed with a commercially available, and FDA approved Dixon T1-weighted based segmentation method. The simultaneously acquired MR imaging protocol is reported in Table 1.

Table 1 –

MR imaging protocol

MR Sequence	Plane	iPat	TR (ms)	TE (ms)	Matrix	NEX	FOV (mm)	Thickness (mm)	Gap (mm)	FA (degrees)	Voxel size (mm)	TI (ms)	Fat saturation
DWI (b-values 50-400-800) without contrast	axial	2	9100–18800	66–83	112*156	2	420	6.0	0.6		2.7*2.7*6.0	220
T1w Dual GE	axial	0	90	1st TE 1.2 2nd TE 2.46	192×256	1	380	5.0	6.0	32°	1.05×1.5×5.0
T2w FSE FS	axial	2	3740	100	206×448	2	400	5.0	6.5		1.3×0.9×5.0		SPAIR
T2w HASTE	coronal	3	1400	66–96	253×256	1	380	5.0	6.0		1.5×1.5×5.0
T1w VIBE Contrast-Enhanced	axial	2	4.06–4.1	1.81–1.91	180×230	1	380	3.0	0	9°	1.6×1.2×3.0		Quick spectral fat saturation

VIBE: volume interpolated breath hold T1 weighted. HASTE: half Fourier single-shot fast spin-echo T2 weighted. FSE: fast spin-echo. FS: fat saturated. GE: gradient echo. iPat: integrated parallel acquisition technique. TR: time of repetition. TE: time of echo. FOV: field of view. FA: flip angle. TI: time of inversion. SPAIR: spectral adiabatic inversion recovery.

Diagnostic contrast-enhanced CT images were acquired using 64 detector spiral Siemens or GE scanners. The images were reconstructed at 2.5 mm slice thickness, with a standard soft tissue reconstruction kernel in the transverse, sagittal and coronal planes. CT scans were acquired within a month before or after the PET/MR and before any treatment.

Tumor volume definition

CT and PET/MR images were transferred to a MIM planning system (MIM software version 6.9.3). An initial automatic rigid registration between CT and MR was performed manually and adjusted if necessary. Regions of interest (ROIs) were drawn by the same radiation oncologist (JW) for all modalities and all patients.

For each patient, gross tumor volume (GTV) of the primary mass (GTV_M) and the lymph nodes (GTV_LN) were first contoured on CT images (${\text{GTV}}_{\text{M}}^{\text{CT}}$ and ${\text{GTV}}_{\text{LN}}^{\text{CT}}$), slice by slice by a radiation oncologist (JW),who was blinded to MR and PET images. Lymph nodes were considered involved if per standard size criteria, their short axis diameter was greater than 1 cm.

Next, the GTV_M and GTV_LN were contoured on the MR images with the same method (${\text{GTV}}_{\text{M}}^{\text{MR}}$ and ${\text{GTV}}_{\text{LN}}^{\text{MR}}$). The CT images were considered when contouring the GTV, but the operator remained blind to the PET images. The lymph nodes were considered involved if their short axis was greater than 1 cm or if they were markedly restricted on the apparent diffusion coefficient (ADC) map.

Lastly, the GTV_M and GTV_LN were contoured visually on the PET/MR fused images with the same method (${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ and ${\text{GTV}}_{\text{LN}}^{\text{PET}/\text{MR}}$). The threshold for contouring lesions using PET was set at the sum of mean standardized uptake value (SUV_mean) and two standard deviations from the normal liver’s mean. CT images were used as a guide when contouring the GTV to exclude uptake in the adjacent structures such as colon. The contours were further adjusted based on the MR information when appropriate. For instance, if part of a lesion was not ¹⁸F-FDG-avid but showed enhancement on MRI, it was added to contour drawn based on PET (Figs. 1 and 2). Lymph nodes were considered involved if they were ¹⁸F-FDG-avid, their short axis was greater than 1 cm on the MR or showed restricted diffusion on the ADC map.

Fig. 1.

Partially ¹⁸F-FDG-avid tumor with a marked enhancement of non-¹⁸F-FDG avid tumor segment detected by MR (a) but not detected on PET (b, c). Positive FDG-avid tumor margin exclusively detected by PET; GTV ^MR (green) and GTV ^PET/MR (red)

Fig. 2.

Additional non-enhancing a ¹⁸F-FDG-avid b, c tumor segment exclusively detected by high FDG uptake on PET; GTV ^MR (green) and GTV ^PET/MR (red)

For each modality, clinical target volume (CTV) was contoured with a 7 to 10 mm margin around the GTV in each direction with MIM’s auto-expansion function, concerning the natural anatomical barriers and the organs at risk.

Spatial coverage analysis

To analyze the difference between volumes, Dice similarity coefficient (DSC) was calculated [28]. DSC measures spatial overlap between two segmentations (A and B target regions) and is defined as DSC (A, B) = 2(A⋂B)/(A + B), where ⋂ is the intersection. A DSC equal to 1 means a perfect overlap between A and B, whereas a DSC equal to 0 means no overlap between A and B.

Statistical analysis

Qualitative variables are presented as absolute and relative frequencies, whereas quantitative variables are presented as median and range. Friedman test and Wilcoxon signed-rank test were used to compare GTVs and CTVs measured by CT, MR, and PET/MR for each tumor or lymph node. A p-value less than 0.05 was considered significant. Statistical analysis was performed using SPSS software (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, Version 20.0. Armonk, NY: IBM Corp.).

Results

Population

We assessed 22 patients with 23 tumors (bifocal tumor in 1 patient) from a total of 93 CC patients who underwent PET/CT at our institute in the study period. Patients’ median age was 68 years (range 44 to 82), with an equal number of male and female subjects. Intrahepatic cholangiocarcinoma (ICC) was the most common tumor subtype with 15 (65.2%) of all tumors. Tumors’ median SUV_max-bw was 5.4 (range 1.4 to 24.3). Patients’ and tumors’ characteristics are presented in Table 2. Eight patients did not undergo diagnostic CT scans within a month from the PET/MR and before treatment. In the remaining fourteen patients, 8 (57.1%) underwent CT on the same day, 5 (35.7%) within 8 days and 2 (14.2%) on 22 and 25 days before PET/MR. The median time between the CT scan and the PET/MR was 1 day (range 0 to 25 days).

Table 2 –

patients’ and tumors’ characteristics

Patients (n=22)
			Median (range)
	Age, years		68 (44 – 82)
			n (%)
	Male sex		11 (50%)
	Lymph node involvement		15 (68.2%)
	Metastases		2 (9.1%) *
Tumors (n=23)			n (%)
	Histology	ICC	15 (65.2%)
		PCC and DCC	6 (26.1%)
		Gallbladder cancer	1 (4.3%)
		Mixed HCC-ICC	1 (4.3%)
	SUVmax-bw		5.4 (1.4 – 24.3) **

^*1 peritoneum involvement; 1 liver + peritoneum involvement

^**3 (13%) tumors were not ¹⁸F-FDG-avid

Quantitative data are presented as median and range.

Qualitative data are presented as absolute and relative values.

ICC: intrahepatic cholangiocarcinoma; PCC: perihilar cholangiocarcinoma; DCC: distal cholangiocarcinoma; HCC: hepatocellular carcinoma

SUV_max-bw: maximum standard uptake value body-weighted

Primary mass volume analysis

CT versus MR comparison

All recorded lesion volumes are presented in table 3. In the 14 patients (15 tumors) with an available diagnostic CT, the ${\text{GTV}}_{\text{M}}^{\text{CT}}$ had a median value of 96 ml (range 4 to 564 ml), and the ${\text{GTV}}_{\text{M}}^{\text{MR}}$ had a median value of 109 ml (range 11 to 635 ml). Although the ${\text{GTV}}_{\text{M}}^{\text{MR}}$ was greater than the ${\text{GTV}}_{\text{M}}^{\text{CT}}$ in 10 cases (mean variation: +32%) and lower in 5 cases (mean variation: −38%), there was no statistically significant difference between these volumes (p value = 0.078). The ${\text{CTV}}_{\text{M}}^{\text{CT}}$ (median: 247 ml, range 27 to 877 ml) and ${\text{CTV}}_{\text{M}}^{\text{MR}}$ (median: 187 ml, range 46 to 957 ml) did not differ significantly either (p value = 0.363).

Table 3 –

GTV and CTV volumes (ml) of primary tumors and lymph nodes, overall and based on tumor location

	Overall population					Intrahepatic tumors					Extrahepatic tumors
	n	mean	SD	median	range	n	mean	SD	median	range	n	mean	SD	median	range
GTVMCT	15	127.3	146.8	96	4–564	10	170.8	161.6	126.5	29–564	5	40.4	50.3	22	4–129
GTVMMR	23	124.7	144.2	69	11–635	16	157.2	157.6	102.5	22–635	7	50.6	70.1	34	11–207
GTVMPET/MR	23	140.3	153.5	94	16–655	16	171	164.2	117.5	22–655	7	70	103.4	37	16–303
CTVMCT	15	261.5	228.8	247	21–877	10	342.4	237.4	298	83–877	5	99.8	86.1	72	21–247
CTVMMR	23	244.4	232.1	171	39–957	16	295.7	253	215.5	47–957	7	127.1	120.3	95	39–380
CTVMPET/MR	23	258.8	230	192	40–987	16	304	248	231	47–987	7	155.7	149.9	117	40–487
GTVLNCT	7*	7.7	5.1	5	4–16	4*	9.8	6.1	9.5	4–16	3*	5	1.7	4	4–7
GTVLNMR	14*	9.8	8	7	1–26	11*	11.3	8.3	8	1–26	3*	4.3	2.5	4	2–7
GTVLNPET/MR	14*	11.1	8.4	8.5	1–27	11*	12	9.1	8	1–27	3*	7.7	5.1	9	2–12
CTVLNCT	7*	87.1	65.1	83	17–204	4*	107.3	71.4	95.5	34–204	3*	60.3	56.3	40	17–124
CTVLNMR	14*	77.7	63.7	48	5–219	11*	70.3	61.4	39	5–219	3*	105	78.2	136	16–163
CTVLNPET/MR	14*	88.8	83.4	49	5–290	11*	78.2	81	39	5–290	3*	127.7	97.4	157	19–207

^*only patients with LN involvement are presented

All volumes are expressed in milliliters

SD = standard deviation

GTV = gross tumor volume; CTV = clinical target volume; M = mass; LN = lymph nodes; CT = computed tomography; MR = Magnetic resonance imaging; PET = positron emission tomography

CT versus PET/MR comparison

Of 15 intrahepatic tumors, in 3 tumors (2 patients) the ¹⁸F-FDG uptake was hypo- or iso-intense to the liver parenchyma on PET/MR images (median SUV_max of 1.42, range 1.72 and 3.75) and did not meet the threshold to be properly separated from the background liver uptake. In those cases, the ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was then considered equal to the ${\text{GTV}}_{\text{M}}^{\text{MR}}$. In all other tumors, lesions were at least partly above the liver background threshold as defined in the methods section and the ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was contoured based on both PET and MRI data. ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was significantly higher than the ${\text{GTV}}_{\text{M}}^{\text{CT}}$ (113 vs 96 ml, range 20–655 vs 4–564 ml, p value = 0.035). The ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was greater than the ${\text{GTV}}_{\text{M}}^{\text{CT}}$ in 11 cases (mean variation + 35%), lower in 3 cases (mean variation − 29%), and unchanged in 1 case. The ${\text{CTV}}_{\text{M}}^{\text{CT}}$ (247 ml, range 27 to 877 ml) and ${\text{CTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ (228 ml, range 78 to 987 ml) did not differ significantly (p value = 0.099).

MR versus PET/MR comparison

In 22 patients (23 tumors) who underwent a PET/MR, there was a statistically significant volume increase of ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ (94 ml, range 16 to 655 ml) compared to ${\text{GTV}}_{\text{M}}^{\text{MR}}$ (69 ml, range 11 to 635 ml) (p value = 0.0001). The ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ was greater than the ${\text{GTV}}_{\text{M}}^{\text{MR}}$ in 17 cases (mean variation + 15%), lower in 2 cases (mean variation − 8%), and unchanged in 4 cases. The ${\text{CTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ (192 ml, range 40 to 987 ml) was also significantly greater than the ${\text{CTV}}_{\text{M}}^{\text{MR}}$ (177 ml, range 39 to 957 ml) (p value = 0.024).

Subgroup analysis based on tumor subtypes

When the tumors were divided into intrahepatic (ICC) and extrahepatic (ECC) subgroups, the ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ remained significantly greater than the ${\text{GTV}}_{\text{M}}^{\text{MR}}$ in both groups (117.5 vs. 102.5 ml, range 22 to 655 vs. 22 to 635 ml, p value = 0.004 and 37 vs. 34 ml, 16 to 303 vs. 11 to 207 ml, p value = 0.042, respectively). However, no statistically significant difference was observed when comparing ${\text{GTV}}_{\text{M}}^{\text{CT}}$ to ${\text{GTV}}_{\text{M}}^{\text{MR}}$ and ${\text{GTV}}_{\text{M}}^{\text{CT}}$ to ${\text{GTV}}_{\text{M}}^{\text{PET}/\text{MR}}$ in either subgroup.

Lymph node volume analysis

Seven of 14 patients (50%) had a lymph node involvement on CT, with a median ${\text{GTV}}_{\text{LN}}^{\text{C}}$ of 5 ml (range 4 to 16 ml). Fourteen of 22 patients (63.6%) had lymph node involvement on MR and PET/MR with a median ${\text{GTV}}_{\text{LN}}^{\text{MR}}$ of 7 ml (range 1 to 26 ml) and a median ${\text{GTV}}_{\text{LN}}^{\text{PET}/\text{MR}}$ of 8 ml (range 1 to 28 ml), respectively. There was a significant difference in volume between the ${\text{GTV}}_{\text{LN}}^{\text{CT}}$ (median 5 ml, range 4 to 16 ml) and the ${\text{GTV}}_{\text{LN}}^{\text{PET}/\text{MR}}$ (median 8.5, range 1 to 27 ml) (p value = 0.026), but no significant difference was found between the ${\text{GTV}}_{\text{LN}}^{\text{CT}}$ and the ${\text{GTV}}_{\text{LN}}^{\text{MR}}$ or between the ${\text{GTV}}_{\text{LN}}^{\text{MR}}$ and the ${\text{GTV}}_{\text{LN}}^{\text{PET}/\text{MR}}$ (p value = 0.074 and 0.394, respectively). Four of the 14 patients (28.5%) who underwent both PET/MR and CT had a lymph node involvement exclusively seen on PET/MR, but not on CT. Lymph nodes were ¹⁸F-FDG-avid in 6 of the 14 cases (42.8%) and lymph node involvement was more extensive on PET/MR than on MR and CT in 3 of those cases.

Primary mass spatial coverage analysis

In the study population, the median Dice similarity coefficient between CT and MR (DSC_CT-MR) was 0.58 (range 0.03 to 0.83). When the tumors were divided into intrahepatic versus extrahepatic subgroups, the DSC_CT-MR was greater in the intrahepatic group (median 0.69, range 0.23 to 0.83) and lower in the extrahepatic group (median 0.41, range 0.03 to 0.74).

The median DSC between CT and PET/MR (DSC_CT-PET/MR) was 0.58 (range 0.22 to 0.87) in the overall population and was greater in the intrahepatic group (median 0.72, range 0.23 to 0.87) and lower in the extrahepatic group (median 0.45, range 0.22 to 0.58).

The median DSC between MR and PET/MR (DSC_MR-PET/MR) was 0.82 (range 0.25 to 1.00) in the overall population and was greater in the intrahepatic group (median 0.84, range 0.76 to 1.00) and lower in the extrahepatic group (median 0.62, range 0.25 to 0.87).

Discussion

PET/MRI has been developed as a hybrid imaging technology to overcome limitations of previously available imaging modalities by combining the structural and molecular information obtained simultaneously from MRI and PET. PET/MR has a lot of advantages over PET/CT including improved lesion detection in solid organs, improved T staging for cancers such as cervical cancer, better motion correction, lower level of ionizing radiation and multiparametric MR imaging. For liver imaging, there are additional technical advantages for performing PET/MR over PET/CT [29]. If a dedicated multiparametric liver MR protocol is performed, this allows additional time for a high-count PET acquisition at the liver bed position which reduces image noise and special resolution of image. In addition, MR imaging allows for respiratory gating of PET data, allowing for decreased motion artifact and improve co-registration of PET and MRI image of the liver [30, 31]. In addition, use of advanced MR imaging techniques to assess tissue perfusion, diffusion, and tumor metabolites, also provides additional dimension of correlative data that is not available with PET/CT [31].

The utility of ¹⁸F-FDG PET/MRI for characterization of liver lesions is well known and is mainly due to the superior tissue contrast of MRI and functional data from PET for detection and characterization of liver lesions. In CC, National Cancer Care Network (NCCN) recommends MRI for diagnosis and staging and ¹⁸F-FDG-PET for locoregional and distant metastases [32, 33]. The combination of ¹⁸F-FDG PET and MRI improves the overall diagnostic accuracy for CC. CCs are most commonly adenocarcinomas which are often ¹⁸F-FDG avid. However, a high proportion of fibrous stroma in some tumors could markedly decrease the ¹⁸F-FDG avidity of CC lesions. In these scenarios, DCE MR shows delayed tumor enhancement due to expanded extracellular space and can better delineate the tumor boundaries. Other potential useful scenario for utilizing PET/MR is to assess the site biliary strictures, where absence of hypermetabolism at the stricture site often indicates a benign etiology. Another potential use for PET/MR in CC is the assessment of resectability. PET/MRI could be advantageous in this setting to evaluate ductal involvement [31, 34, 35].

In this study, we assessed tumor and lymph node delineation using ¹⁸F-FDG PET/MR compared to CT and MR in untreated CC patients for the first time. Our findings suggest that delineation based on ¹⁸F-FDG PET/MR images yields significantly higher tumor GTVs than both CT and MR, conventional modalities used for treatment planning in CC. This is in concordance with previous studies that used ¹⁸F-FDG PET/CT data alongside MR in RT planning of colorectal cancer liver metastases and demonstrated an increase in GTVs compared to using CT alone, without significant change in radiation to OAR [21]. Although the importance of more accurate tumor delineation and larger GTV in the CC has not been extensively studied, there is evidence in other cancer types, such as pelvic or head and neck cancers, that an increase in GTV results in improved outcomes by better tumor delineation [36–38]. Our study showed that ¹⁸F-FDG PET and MR imaging provide complementary information which translates in better tumor delineation than either modality alone. In fact, the combination of both imaging technologies eliminates the shortcoming of individual imaging modalities in estimating tumor volume in case of hypometabolic lesions or heterogeneous tumor uptake in ¹⁸F-FDG PET and lack of enhancement in dynamic contrast-enhanced (DCE) MR. In our study, the use of PET/MR resulted in significantly greater median GTVs using than MR in both ICCs and ECCs. We observed no significant difference between MR- and CT-based GTVs, despite expected MR advantage due to better soft tissue contrast. This observation suggests that, at least in our patients, MR alone provided little added value to CT-based delineation, which is the standard of care. In fact, some studies suggest that MR tends to underestimate CC tumor size compared to pathologic specimens [39].

¹⁸F-FDG PET/MR also performed better than CT and MR for delineation of involved lymph nodes. PET/MR yielded significantly higher GTV_LN compared to CT (p value = 0.026) but not MR (p value = 0.394). However, we observed a higher number of involved nodes on PET/MR compared to CT and MR alone in approximately 25% of patients, as in these patients, the nodes did not meet the size criterion for pathologic involvement while they were intensely ¹⁸F-FDG avid. This is in concordance with the findings of multiple studies that showed a higher detection rate of nodal involvement by PET compared to MR and CT [20, 40, 41]. Although we did not directly compare PET/MR and PET/CT for detection of nodal involvement in CC, we believe that higher soft tissue contrast of PET/MR and the use of additional sequences such as diffusion-weighted imaging (DWI) could improve the detection of nodal involvement in PET/MR compared to PET/CT. In fact, we noted that 57.2% of the nodes defined as involved by PET/MR were not ¹⁸F-FDG avid. This points out PET and MR’s complementary role in identifying the involved nodal disease and potentially improved radiation field planning compared to PET/CT.

The median DSC was 82% between PET/MR and MR GTV, while it was similar and significantly lower between PET/MR and CT (58%) as well as MR and CT (58%). This was expected as the PET/MR GTV was always inclusive of MR only GTV, while there was no specific overlap between the PET/MR data and CT [42]. The 18% dissimilarity of the PET/MR and MR GTV is due to the contribution of the ¹⁸F-FDG PET to a larger GTV in PET/MR compared to MR, including the non-overlapping tumor mass defined by PET, which reflects an advantage of PET/MR for inclusion of non-enhancing but ¹⁸F-FDG-avid segments of tumors. This could further optimize the RT planning and radiation field, although the effect of this change in the patients’ overall outcome was not investigated in this study. DSC_PET/MR-MR was higher in ICCs compared to ECCs, which is likely due to a higher rate of tumors that were iso- or hypometabolic relative to the liver parenchyma in the ICC subgroup compared to ECC and greater heterogeneity of ¹⁸F-FDG uptake in ECCs compared to ICCs [18, 43, 44].

Our study had several limitations. Given this study’s retrospective design, we could only compare the GTV of the tumors defined by PET/MR and other modalities. Therefore, while we showed that PET/MR delineated overall a larger area for radiation than other modalities, we could not investigate the effect of this change on the patients’ outcome. We believe that our study has adequately shown the merit of this concept and its potential to justify designing future controlled clinical trials to precisely assess the impact of improving tumor delineation using PET/MR in CC on outcomes. We also note that the number of patients in this study was relatively low due to the limited number of CC patients who undergo PET for RT planning. Over the 3-year period that we collected the patients, only 93 patients underwent PET/CT and of those only 22 were additionally imaged with PET/MR. This may be in part due to the fact that no prior study has investigated the usefulness of PET/MR in CC for improving RT planning. Lastly, in a cohort of our patients, we had to exclude a few patients from some of the analysis since they did not undergo all the imaging modalities within the defined timeframes for inclusion in the analysis, which is mostly due to the retrospective nature of this study. We hope to address these shortcomings in the future by meticulously designed controlled prospective studies.

In conclusion, simultaneous metabolic imaging using PET and high contrast anatomic imaging using MR are complimentary for improving tumor and lymph node delineation in CC, making ¹⁸F-FDG PET/MR a potentially valuable tool for RT planning in these patients. Some of the advantages of PET/MR over other modalities include but are not limited to the proper fusion of PET and MR, allowing better margin and LN detection, lower radiation exposure to the patient, and potentially decreasing the number of scans the patient would need to undergo for RT planning. Given the encouraging results of this study, future studies should focus on patient outcomes and cost-effectiveness analysis to help incorporate this modality in the everyday practice of hepatic radiation oncology centers.