Automated Radiotherapy Planning Using Artificial Intelligence: A Comparative Study of Dose Distribution Quality in Head and Neck Cancers

Mahmoud A

Published on: 2025-06-14

Abstract

This study evaluated three artificial intelligence-based automated radiotherapy planning approaches for head and neck cancers against conventional manual planning methods. A retrospective cohort of 150 head and neck cancer patients treated with IMRT/VMAT (2020-2024) was analyzed using Knowledge-Based Planning (KBP), Deep Learning-based Dose Prediction (DL-DP) with hierarchically densely connected U-Net architecture, and Generative Adversarial Network-based Planning (GAN-P). Comprehensive dosimetric evaluation included planning target volume coverage metrics, organ-at-risk dose analysis, conformity indices, and three-dimensional gamma analysis. All automated approaches achieved equivalent target coverage to clinical plans (V95%: 96.2-97.1% vs. 96.8±2.1%, p=0.23) while demonstrating significant improvements in organ sparing, particularly for parotid glands with mean dose reductions of 3.3-4.7 Gy (all p<0.01). Xerostomia normal tissue complication probability was reduced from 23.4±8.7% to 18.2-20.8% across automated methods (p<0.05). Gamma analysis showed excellent agreement with pass rates >95% for 3%/3mm criteria. Clinical acceptability rates were high for all methods (81.3-93.3% vs. 89.3% for clinical plans). Planning time was dramatically reduced from 4.2±1.8 hours to 12-23 minutes, representing 93-95% efficiency improvements. These findings demonstrate that AI-based automated planning achieves dosimetric quality equivalent to or superior to manual planning while providing substantial improvements in planning efficiency and normal tissue sparing. The clinically meaningful reductions in parotid doses and xerostomia risk, combined with high clinical acceptability and dramatic efficiency gains, support the readiness of automated planning systems for widespread clinical implementation in head and neck cancer radiotherapy.