Speakers
Description
Purpose
The clinical target volume (CTV) is a weak link in the radiotherapy chain. In head and neck cancer it has two parts, and both are drawn as binary contours. The primary CTV is a margin expansion of the GTV, edited at anatomical barriers. The elective nodal CTV includes or leaves out whole lymph node levels. Yet the risk of microscopic disease varies continuously across voxels and depends on the patient’s anatomy and clinical presentation. Models exist for each part separately: hidden Markov models of lymphatic progression, and Eikonal-based models of tumor spread that respect anatomical barriers. They have not yet been combined. We propose a clinical target map (CTM) of voxel-wise disease probability that brings both together, and we use it as voxel weights in probabilistic plan optimization.
Methods
Head and neck cases from the TROTS dataset were used, with CT, GTV, clinical CTVs, OAR contours and dose-influence matrices. CT numbers were used to sort soft tissue into fat, muscle and mucosa. Bone, cartilage and air were treated as impermeable barriers. Microscopic primary extension was modeled as a front moving out from the GTV surface, governed by the Eikonal equation and solved by fast marching. In each scenario, the tissue resistances, the margin, the chance of a barrier breach and the barrier contour uncertainty were all sampled. In the same scenario, a nodal involvement pattern for levels I–V and VII was drawn jointly from the posterior of a lymphatic progression model. Drawing the pattern jointly keeps the correlations between levels. The CTM was the voxel-wise fraction of N scenarios containing each voxel. A plan with elective under-dosage penalties weighted by the CTM was optimized in OpenTPS. It was compared with a reference plan built on the binary clinical CTVs, using identical beams and OAR objectives. The comparison covered DVHs, OAR dose metrics, probability-weighted target coverage and expected TCP.
Results
The CTM showed spatially graded disease probability instead of uniform inclusion or exclusion. Probability was shaped by anatomical barriers around the primary tumor and graded across lymph node levels. The CTM-weighted and reference plans were compared in terms of target coverage, organ-at-risk sparing and expected TCP. Differences between the two plans were concentrated in regions where the CTM and the clinical CTVs disagreed. Quantitative comparisons across all cases are ongoing.
Conclusion
Combining anatomy-constrained primary spread with lymphatic progression gives voxel-wise probabilistic target for head and neck cancer, built from routinely available data. Used as optimization weights, it changes how dose is distributed across elective regions compared with binary CTV-based planning. This supports its further investigation as an alternative to binary CTVs in probabilistic planning. Future work includes more patients, calibration against pathology and/or recurrence data, and robust evaluation.
Acknowledgement: This work is part of the UPLIFT project and funded by the European Union under Grant Agreement No. 101168955.