Dysphagia optimized knowledge-based planning for head and neck cancer

Swallowing dysfunction after radiotherapy (RT) is often linked to pharyngeal mucosal damage. This study aimed to develop a dysphagia-optimized knowledge-based planning (DO-KBP) model by incorporating individual pharyngeal constrictors (DO-KBP) into an existing model, which was based on a conventional approach of including pharynx as a single structure during treatment planning (P-KBP).

The P-KBP model was trained on 175 head and neck cases with the pharynx contoured as a single organ. The DO-KBP model included 36 additional oropharynx cases (∼20% increase) with individual pharyngeal constrictors delineated. Both models were evaluated on 25 test patients. Treatment plans generated by each model were normalized to planning-target-volume (PTV) coverage (D95%), and dosimetric parameters were compared using two-tailed paired t-tests. A blind physician review assessed clinical preference.

The DO-KBP model was able to significantly reduce the mean dose to the inferior constrictor (36.52 ± 9.87 Gy to 19.52 ± 6.23 Gy) and superior/middle constrictors (51.89 ± 6.31 Gy to 47.46 ± 6.12 Gy) (p < 0.05) with the addition of 36 high-quality treatment plans. Though statistically significant increases in mean dose were observed for the spinal cord PRV (D0.03cc), cochlea, mandible, and left brachial plexus with the DO-KBP model, these differences were small in magnitude and remained within clinical goals. However, these differences were small in magnitude and remained within clinical goals. Plan homogeneity was equivalent (HI = 0.09). The DO-KBP plans were preferred in blinded review.

Targeted addition of a small number of cases with individually contoured constrictors to an existing model significantly improved sparing of swallowing structures, without compromising overall plan quality or increasing organs-at-risk (OAR) doses beyond clinical thresholds, highlighting that modest, focused data augmentation can yield clinically meaningful gains.