Initial Observation of Contrast Profiles for 2D and 3D MRI Sequences in MR-Guided Radiation Therapy for Locally Advanced Pancreatic Cancer

Neoadjuvant Therapy for Pancreatic Cancer

Pancreatic cancer (PC) is a highly lethal malignancy with a 5-year survival rate of 11%.¹ At diagnosis, nearly 50% of patients present with locally advanced pancreatic cancer (LAPC). For PC as a whole, an R0 resection seems to be the greatest positive factor affecting outcomes. While systemic treatment of PC has improved, surgical resection rates remain low, and patients with borderline resectable or unresectable LAPC still have poor outcomes.

It is hypothesized that neoadjuvant treatment can improve rates of R0 resection, local control and survival²; however, there have been conflicting results regarding the optimal neoadjuvant regimen. A phase 3 trial failed to demonstrate a benefit with the addition of neoadjuvant conventionally fractionated RT (CFRT) to chemotherapy following induction chemotherapy.³ However, a propensity-matched retrospective review comparing neoadjuvant stereotactic body radiation therapy (SBRT) to CFRT demonstrated an improved overall survival with SBRT.⁴ Other studies have demonstrated higher biological effective dose (BED) with SBRT as a predictor of improved overall survival (OS).⁵ A meta-analysis of over 1000 patients across 19 trials concluded that SBRT did not have superior outcomes compared with conventional RT, but they postulated that dose escalation would likely have a clinical benefit.⁶ This echoes the sentiment of other studies regarding the dosimetric feasibility of dose-escalating CT-based SBRT for LAPC on linacs.⁷ Unfortunately, the success of neoadjuvant CT-based SBRT for LAPC has been limited due to excess dose resulting in toxicity to the OARs and use of non-ablative doses that did not clearly demonstrate improved OS for inoperable PC.^8,9

MR-Guided Adaptive SBRT

MR-guided RT (MRgRT) is a novel approach utilizing MRI scans, acquired both prior to and continuously during treatment to enable soft tissue patient positioning, beam gating based on soft tissue tracking, and online adaptive planning. Stereotactic magnetic resonance-guided adaptive radiation therapy (SMART) utilizes the unique aspects of MRgRT mentioned above to push radiation target doses higher while respecting established OAR constraints.

Preliminary data of SMART for LAPC from a single institution demonstrated grade 3 toxicities at only a ∼3% rate with a median follow up of 10.3 months.¹⁰ A multicenter phase II trial, which recently closed to accrual, implemented SMART in an effort to increase BED to 100 Gy, using an integrated 0.35T MR-cobalt or MR-linac system (ViewRay Inc., MRIdian®, Denver, CO, USA).¹¹ Details regarding the MR-cobalt and MR-linac technologies and associated workflow have previously been reported.^12,13 The GTV was defined as the tumor. The CTV was defined as GTV with or without regional lymph nodes and a 3 mm margin was used for PTV construction. The prescription dose was 50.0 Gy in 5 fractions.

Per protocol, prior to every treatment, a volumetric MRI scan (3D MRI) was performed to visualize anatomic changes and predict radiation dose to both target volumes and OARs as per the original plan. When OAR constraints were violated because of anatomic changes, the study required that on-table adaptive planning be performed to ensure OAR constraints were met; optimizing target volume coverage was a secondary objective. The original plan was not delivered if: 1) a GI OAR constraint was violated 2) coverage of the CTV was <85% by the 47.5 Gy isodose line 3) there was a favorable shift in CTV and OAR doses in the adaptive plan such that the CTV coverage by the 47.5 Gy isodose line is improved by 10% or more compared with the original plan. On-table plan adaptation, plan evaluation, and adaptive quality assurance occurred while the patient remained in the treatment position. During treatment delivery, sagittal planar MR images (2D cine) were continuously acquired during treatment delivery at either 4 or 8 frames per second and the radiation beam was automatically paused when the tracked pancreatic tumor moved, typically due to respiratory motion, out of a defined gating boundary.

MR Sequences

True fast imaging with steady-state free precession (TRUFI) is a T1/T2 weighted MRI sequence which utilizes steady states of magnetizations through gradient reversal echoes and short repetition time between successive excitation pulses thereby offering high signal-to-noise ratio per unit time.¹⁴ Work has been done for decades to improve these methods and preserve high signal-to-noise in the presence of motion.¹⁵ An application in radiation therapy of TRUFI is “real time” tracking that is employed in MRgRT by observing motion on a 2D sagittal slice.

Due to the different sequences to visualize the 2D real time cine and the initial pretreatment 3D setup MRI, there is a large discrepancy in the intensity of these structures. Based on this background, we aimed at devising a qualitative guide to help establish a rule of thumb for physicians using the MRgRT system.

Accessing Image Sets

With IRB approval, DICOM files for both 2D cine images and 3D MRI images stored in the treatment planning system (TPS) were retrospectively accessed for patients with PC (n=20) treated on the MR-linac. At the time of simulation, an appropriate tracking target was identified and contoured on a single 3mm sagittal slice of the 3D MRI. At our institution, the default structure selected for tracking is often the visible tumor.

This sagittal slice from the 3D MRI was then directly compared to the co-registered 7mm 2D cine to identify structures with notable discrepancies in signal intensity paying particular interest to arteries, veins, and biliary structures. The 3D MRI was then explored in additional planes to confirm structure identities. This process is summarized in Figure 1.

• Download figure
• Open in new tab

Figure 1.

Schema

Schema of our methodology for reviewing and comparing the contrast profiles for structures on 2D cine and 3D MRI scans.

Qualitative Description

We qualitatively identified the relative intensity of arteries (specifically the aorta, celiac axis, and superior mesenteric artery (SMA)), veins (portal vein and superior mesenteric vein (SMV)), and biliary radicles (common bile duct (CBD) and pancreatic duct (PD)).

The relative intensity of each structure was compared between sequences. Any structure that was bright was labeled hyperintense; hyperintensity was subdivided into starkly hyperintense and mildly hyperintense. Any structure that was dark was labeled hypointense. Any structure with similar intensity to surrounding structures or tissues was labeled isointense.

Quantitative Verification

A custom Python script was used to read the raw image data from MRI DICOM files using the open source PyDicom module (citation: https://zenodo.org/record/5543955#.Y0nnUXbMLHI). Prior to intensity distribution comparison, the 3D MRI volume with in plane resolution of 1.5 × 1.5 mm was resampled to match the in plane resolution of the cine scan (2.4 × 2.4 mm), and the 2D slice closest to the axial location as the cine scan was extracted from each 3D volume. Additionally, a single frame in the deep breath hold position was sampled for comparison to the 3D scan. Pixel intensities values across all images were concatenated to form the two samples: MRI setup scan and MRI cine scan intensities. These samples were then compared using two sample Kolmogorov-Smirnov test available from SciPy (https://scipy.org/citing-scipy/, scipy.stats.ks_2samp — SciPy v1.7.1 Manual). Additionally, the intensity distributions were visualized using a histogram plot generated using Matplotlib (https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.hist.html).

Case 1

Our first example involves a female patient in her mid 60s who initially presented with weight loss and jaundice. Workup led to diagnosis of a T4N0M0, Stage III (AJCC 8^th edition) unresectable adenocarcinoma of the pancreatic head. She underwent 7 cycles of neoadjuvant chemotherapy (4 cycles of FOLFIRINOX, switched to 3 cycles of gemcitabine/paclitaxel due to inadequate response). Restaging workup demonstrated a decreased CA-19-9 and a partial response to chemotherapy on MRI. She then underwent preoperative SMART to the pancreatic head tumor in 50 Gray in 5 fractions. She was slated for pancreaticoduodenectomy which was aborted due to inability to separate the tumor from the common hepatic artery or the replaced right hepatic artery originating from the superior mesenteric artery. The replaced right hepatic artery is indeed visualized in Figure 2.

• Download figure
• Open in new tab

Figure 2.

Case 1 Images

Comparison of sagittal slices of the 2D cine (panel A) to the corresponding 3D MRI (panel B) for Case 1.

Figure 2 illustrates the qualitative differences in contrast profiles between the 2D cine (panel A) and the 3D MRI (panel B) observed in her case. In both panels, the tumor is seen contoured in red. Additionally, the yellow contour in panel A represents an expansion boundary used for MRgRT beam gating. Superior to the tumor, in both panels, the confluence of the pancreatic duct and the common bile duct is (PD/CBD; light green label) observed as hyperintense; however, in panel B, it is seen as starkly hyperintense. Immediately superior to the PD/CDB confluence, the portal vein (PV; blue label) is seen hyperintense on the 2D cine yet it appears isointense on the 3D MRI. Posterior and anterior to the PV, respectively, the right and left branches of the hepatic artery (R HA, L HA, red) are visualized with similar intensity on both scans.

Case 2

Our next case is a male in his early 70s who initially presented with dyspepsia, abdominal pain, and weight loss. Workup led to diagnosis of a T4N0M0, Stage III (AJCC 8^th edition) borderline resectable ∼3 cm adenocarcinoma of the pancreatic head with involvement of the superior mesenteric artery. He underwent 8 cycles of neoadjuvant mFOLFIRINOX. Restaging workup demonstrated good response via CA-19-9 and CT scan of the abdomen. He subsequently completed preoperative SMART to the pancreatic head tumor in 50 Gray in 5 fractions. He successfully underwent pancreaticoduodenectomy with pathology demonstrating residual ductal adenocarcinoma of the pancreatic head and metastases in 3 of 12 lymph nodes examined, ypT2N1.

His anatomy can be visualized in Figure 3. In both panels, the tumor is seen contoured in red. As in case 1, the yellow contour in panel A represents an expansion boundary used for MRgRT beam gating. Superior to the tumor, in both panels, the superior mesenteric vein is (SMV; blue label) observed as hyperintense; however, in panel B, it is seen as isointense. Immediately anterior to the SMV, the pancreatic duct (PD; light green label) is seen as hyperintense on the 2D cine, yet, it appears starkly hyperintense on the 3D MRI. Superior to the SMV, the common hepatic artery (C HA; red) is visualized to be hyperintense in both scans. Inferior to the tumor, the duodenum can be visualized in both scans (D; orange), and, inferior to the D, the Aorta (AO; red) can be visualized as hyperintense in both scans.

• Download figure
• Open in new tab

Figure 3.

Case 2 Images

Comparison of sagittal slices of the 2D cine (panel A) to the corresponding 3D MRI (panel B) for Case 2.

Case 3

Our final case is a male in his late 60s who initially presented with abdominal pain and jaundice. Workup led to diagnosis of a T2N0M0, Stage II (AJCC 8^th edition) borderline resectable 3.5 cm adenocarcinoma of the pancreatic head. He underwent biliary stenting followed by 7 cycles of neoadjuvant mFOLFIRINOX. Restaging workup demonstrated good response via CA-19-9 and CT of the abdomen. He subsequently completed preoperative SMART to the pancreatic head tumor in 50 Gray in 5 fractions. He underwent pancreaticoduodenectomy successfully with pathology demonstrating residual ductal adenocarcinoma of the pancreatic head, ypT2N0.

His anatomy can be visualized in Figure 4. In both panels, the tumor is seen contoured in red. As in our prior cases, the yellow contour in panel A represents an expansion boundary used for MRgRT beam gating. Superior to the tumor, in both panels, the common bile duct is (BD; light green label) observed as hyperintense; however, in panel B, it is seen as starkly hyperintense. Immediately superior and posterior to the BD, the portal vein (PV; blue label) is seen hyperintense on the 2D cine yet it appears isointense on the 3D MRI. Superior to the PV, the common hepatic artery (HA; red) is visualized to be mildly hyperintense in both scans. Anterior to the tumor, the duodenum is visualized in both scans (D; orange). Posterior to the tumor, the inferior vena cava (IVC; blue) is visualized as hyperintense on the 2D cine yet isointense on the 3D MRI.

• Download figure
• Open in new tab

Figure 4.

Case 3 Images

Comparison of sagittal slices of the 2D cine (panel A) to the corresponding 3D MRI (panel B) for Case 3.

Qualitative Findings

In all 20 patients reviewed, arteries (aorta, celiac, SMA, HA) typically appeared with mild hyperintensity resulting in similar contrast profiles in both scans. However, veins (PV, SMV) appeared consistently hyperintense on 2D cine and consistently isointense on 3D MRI. Biliary structures (CBD, PD) appeared consistently hyperintense on both scans, only mildly hyperintense on 2D cine but starkly hyperintense on 3D MRI. These findings are summarized in Table 1.

Quantitative Verification

The pixel intensity distributions extracted from 2D cine and 3D MRI images were confirmed to differ significantly (two sample Kolmogorov-Smirnov test; test statistic=0.40; p<0.001) in Figure 5.

• Download figure
• Open in new tab

Figure 5.

Pixel Intensity Histogram

Comparison of Pixel intensity distributions for 2D cine and 3D MRI images.