Related Experiment Video
Updated: May 29, 2026
Proton Therapy Delivery and Its Clinical Application in Select Solid Tumor Malignancies
Published on: February 6, 2019
A coverage probability based method to estimate patient-specific small bowel planning volumes for use in radiotherapy
Liv Bolstad Hysing1, Matthias Söhn, Ludvig P Muren
1Section for Biomedical Physics, University Hospital for Radiation Oncology, Tübingen, Germany. liv.bolstad.hysing@gmail.com
Background And Purpose:
The aim of this work was to develop a statistical method for generation of patient-specific planning organ-at-risk volumes (PRVs) for the small bowel (SB), by efficient use of a few repeat CT scans.
Materials And Methods:
The PRVs are generated from a coverage probability (CP) matrix of the small bowel wall (SBW) by thresholding. To estimate the CPs, we extend a previously published 'relative frequency of coverage' approach by adding a 'soft margin' around each SBW instance. This prevents the CP matrix from containing any holes, thus making it more robust. As the number of CTs approach infinity, the 'soft margin' approaches zero and the CP matrix converges to the 'relative frequency of coverage'. The PRVs were evaluated by using the bootstrap method in three patients with different degrees of SB motion: The PRVs from randomly sampled subsets of CTs were compared to the PRVs generated from all 10-11 CT scans, by analysis of sensitivity and specificity. Furthermore, the PRVs generated for CP=0.005 (i.e. generous patient-specific PRVs) and for CP=0.03 (i.e. tight patient-specific PRVs) were compared to an intestinal cavity (IC) approach and a population based PRV approach of 10 and 30 mm isotropic planning margins around SB.
Results:
The sensitivity and specificity of the PRVs depend on the number of CT scans and the CP threshold. With three CT scans and a threshold of 0.03, an average sensitivity of 94-96% and specificity of 86-97% was obtained. All investigated SB planning volumes had an average overlap >89% of both SB and SBW. The tight patient-specific PRVs and the 10mm margins had the lowest relative volumes, followed by the generous patient-specific PRVs, the 30 mm margins and the ICs.
Conclusions:
Based on a few CTs, our method generates patient-specific SB PRVs which are both sensitive and specific. Compared to conventional approaches, the patient-specific PRVs are either similar or better in predicting for SB voxels, and at the same time they occupy a smaller or similar volume in the patient.
