AI-Driven Oncology Treatment Planning

Abstract

Oncology treatment planning integrates tumour biology, patient physiology, organ-at-risk tolerances, and rapidly evolving therapeutic options into decisions that determine survival and quality of life for cancer patients. The complexity of this integration has grown substantially with the advent of precision oncology, immunotherapy, and multimodal treatment combinations that create a decision space too large for unaided clinical judgement to navigate optimally. AI-driven treatment planning tools offer systematic optimisation across this decision space, from automated radiotherapy contouring and dose optimisation through ML-based systemic therapy sequencing and immunotherapy response prediction. We evaluated five AI treatment planning frameworks applied to 54,286 oncology patients at 18 European cancer centres in Berlin, Vienna, Paris, Stockholm, and Milan between 2017 and 2024, covering non-small cell lung cancer, breast cancer, rectal cancer, and head and neck cancer. Frameworks ranged from AI-assisted contouring in radiotherapy through fully integrated multimodal AI treatment planning with adaptive response-guided modification. Performance was benchmarked using our Oncology Treatment Planning AI Score (OTPAS), integrating plan quality, treatment delivery precision, tumour response rate, toxicity reduction, and clinical outcome improvement. The adaptive AI-integrated framework achieved the highest OTPAS (0.931), improving locoregional control at 2 years from 68.4% (standard planning) to 84.2% and reducing grade 3+ toxicity from 22.4% to 14.6%. Adaptive radiotherapy guided by mid-treatment MRI response showed the strongest tumour control advantage, while AI systemic therapy sequencing produced the largest quality-of-life benefit through toxicity-optimised regimen selection.