Session Item

There are no dose-response models establishing relationships between colorectal doses or dose-volume metrics and late colorectal sequelae (such as subsequent malignant neoplasms) in childhood cancer survivors. Such models do not exist because these studies require large cohorts with decades of follow-up. Consequently, these cohorts are largely comprised of patients treated in the pre-CT era of radiation therapy (RT) where organ dose calculations were not possible. Thus, it is common practice in late effects studies to reconstruct survivors’ RT on computational phantoms to estimate organ doses. However, the Late Effects Group computational phantom, which has been used for hundreds of late effects studies over several decades, does not have a colorectal model. Here, we aimed to (1) add a colorectal model that incorporates pediatric anatomical variations and (2) validate the geometric and dosimetric accuracy of the model across the typical age range of pediatric RT patients.

Whole-body non-contrast CT scans of 114 pediatric patients (age range: 2.1-21.6 years, 74 males, 40 females) were retrospectively selected. Manual colorectal contours were reviewed and approved by two radiation oncologists. 1 patient was used for the anatomical template, 103 for training and 10 for testing. All contours were normalized using median colorectal length and registered to an anatomical template using the constrained symmetric thin-plate spline robust point matching method. Deformed contours were used to create a principal component analysis-based colorectal statistical shape model to extract the dominant deformations of the population. Geometric accuracy was validated using the Dice similarity coefficient (DSC), distance-to-agreement (DTA), and Hausdorff distance (HD) between the patient-specific and model contours for 10 test patients. Dosimetric accuracy was validated using standard Wilms’ tumor 3D conformal RT plans on the test patients. We computed colorectal RT dose data (mean, max, V₅, V₁₀, V₁₅, V₂₀, D₁, D₅₀, and D₉₅) and compared the patient-specific data with that of the colorectal model (Table 1). Workflow illustrated in Fig. 1.

Using the colorectal model on the unseen 10 test patients, the mean (min-max) DSC, DTA and HD between the patient-specific and reconstructed model contours was 0.89 (0.85-0.91), 2.1mm (1.7-3.0), and 8.6mm (5.7-14.3), respectively. For the Wilms’ RT plans, the average absolute difference in mean and max dose, was 0.02Gy (0.00-0.11) and 0.03Gy (0.00-0.21), respectively. For V₅, V₁₀, V₁₅, and V₂₀ all absolute differences were within 1.6%. For D₁, D₅₀, and D₉₅ all absolute differences were within 2.9Gy.

We demonstrated that our colorectal statistical shape model can reconstruct unseen shapes with good accuracy and be used for accurate dose reconstruction. The model will be integrated (Fig. 1F-I) into the Late Effects Group computational phantom and be used to reconstruct colorectal doses for studies of RT-related colorectal late effects.