Introduction:

Chemoradiation therapy (CRT), followed by total mesorectal excision (TME), is the standard-of-care treatment for locally-advanced colorectal cancer. Currently, the tumor delineation in radiation therapy is mostly relied on CT images. With the advancement of fast acquisition of MRI scans and the possibility to obtain reliable electron density, MRI-guided treatment planning and MRI-Linac are gradually gaining its popularity in radiotherapy. Although MRI shows much better soft tissue contrast which could lead to better lesion delineation, the appearance of the tumor can vary greatly depending on the type of imaging sequence or images acquired at different times after injection of contrast agent. Furthermore, patient motion can be another source of variation that can add to the uncertainty when determining the target volume for treatment planning. In this study, we developed a quantitative method using the radial distance, defined as the distance from the center of rectum to the boundary of the tumor region, to evaluate variations of the tumor ROI. All patients had a follow-up MRI, and variations in both pre- and post- treatment MRI were studied.

Methods:

A total of 45 patients (mean age 60) with stage T3 and T4 rectal cancer were studied. The MRI was done on a 3.0 Tesla scanner (GE Signa HDxt) using a phased-array body coil. In this study only the three post-contrast frames (termed as L2, L3, and L4) acquired at 15, 60, and 120 seconds after injection of Gd (0.1 mmol/kg) in the LAVA sequence were analyzed. On each set of DCE images, the tumor region of interest (ROI) was carefully drawn by a radiation oncologist (Figure 1). Using the rectum contour drawn on one slice as reference, an intensity based non-rigid Demons algorithm was applied to automatically segment the rectum in both cranial and caudal directions (Figure 2). Then the centroid of the rectum on each slice was used as the reference point to calculate the radial distance (Figure 3). A histogram was generated for the difference of the measured radial distance between L2 vs. L3, and another for L2 vs. L4. The values of the 25th, 50th, 75th, and 90th percentile pixel, and the mean value averaged over the entire tumor were obtained. In order to evaluate the difference coming from intra-fractionation patient motion, Affine registration was used to co-register two sets of images to determine the deformation matrix, e.g. L2 and L3 (Figure 4), and then the manual ROI drawn on L2 was mapped to L3. For these two ROI’s the difference was coming from the motion.

Results:

Figure 4 demonstrates the difference between the manually drawn ROI on L2 and L3, the mapped ROI from L2 to L3, as well as the histogram of the difference in the radial distance in these two comparisons. Table 1 summarizes the difference in radial distance (absolute value) of the 25th, 50th, 75th, and 90th percentile pixel. The intra-fractionation patient motion was found to be within 1-2 mm. Although no-special bowel preparation was given, the patients were able to stay still during the whole scan. On the other hand, when lesions were defined using different image sequences, over 25% pixels showed over 3-5 mm differences, and over 10% pixels showed 6-8 mm differences.

Conclusions:

The mean value from the entire tumor showed that the difference between L2 and L4 was higher than between L2 and L3, which was reasonable due to the longer time difference. The motion difference was higher in the post-treatment compared to pre-treatment. This was also reasonable as patients might be weaker after treatment and had more difficulty holding still during the scan. In the current CRT planning, 1 cm margin expansion is typically used to create planning target volume (PTV) from clinical target volume (CTV) which counts for both intra- and inter-fractionation motion and setup error. Previous studies assessed the inter-fractionation motion of rectal cancer patients was by average of 8 mm using daily CBCT, and concluded the 1 cm expansion should be adequate for radiotherapy planning [1-2]. However, none of previous work tried to identify the intra-fractionation motion or the variation of CTV delineation from image sequences. This is understandable as in CT based planning, different image sequences as helical mode or cine mode etc., theoretically do not cause large variations in the image contrast. But in MRI-based planning, difference in image contrast as demonstrated in this work, can lead to variations in identifying CTV. With MRI-guided radiotherapy gradually becoming available, the method developed in this work may be used to evaluate the variations in delineating tumor ROI, so a more precise treatment planning field with reasonable margin can be used to avoid side effect without compromising the tumor treatment efficacy.

Acknowledgements

This study is supported in part by NIH R01 CA127927, and Rutgers-RBHS precision medicine pilot grant, The Rutgers-Cancer Institute of New Jersey P30 CA072720.

References

[1]. Yamashita H, Takenaka R, Sakumi A, Haga A, Otomo K, Nakagawa K. Analysis of motion of the rectum during preoperative intensity modulated radiation therapy for rectal cancer using cone-beam computed tomography. Radiat Oncol. 2015;10:2. [2]. Chong I, Hawkins M, Hansen V, Thomas K, McNair H, O'Neill B, Aitken A, Tait D. Quantification of organ motion during chemoradiotherapy of rectal cancer using cone-beam computed tomography. Int J Radiat Oncol Biol Phys. 2011;81(4):e431-8.