Automated daily adaptation for breast cancer radiotherapy: benefits of deep learning-driven dose prediction workflow for CBCT-based adaptation

<i>Objective.</i>Daily anatomical variations can jeopardize the quality of modern radiotherapy. Adaptive radiotherapy (ART) aims to address this by adjusting treatment plans according to the patient's daily anatomy. However, as implementation of ART is currently time-consuming, laborious, and expensive, its routine clinical use remains limited. In this study, we investigate the feasibility of using deep learning (DL)-based dose prediction with automated treatment planning and daily cone-beam CTs (CBCTs) to evaluate the clinical benefits of daily replanning in breast cancer RT.<i>Approach.</i>Dataset of 28 breast cancer patients treated with Volumetric Modulated Arc Therapy and their daily CBCTs (<i>n</i>= 15) were used to evaluate the dosimetric impact of daily anatomical variations. In-house developed DL-based dose prediction engine and automated planning solution were used to create dose accumulated and adaptive plans to determine the potential benefits of daily replanning.<i>Main results.</i>Adaptation improved the planning target volume (PTV) dose coverage in 25 of 28 patients and increased the fractions achieving the clinical goal (PTV<i>V</i>_{38.05 Gy}> 95%) from 35% to 93%. Moreover, ipsilateral lung<i>V₁</i>_{6 Gy}, heart<i>V</i>_{4 Gy}, and contralateral lung<i>V₂</i>_{.5 Gy}decreased by 0.4 pp, 0.6 pp, and 0.2 pp, respectively. Using an adaptation criterion based on DL prediction, approximately 25% of fractions were identified for replanning, resulting in a 1.2 pp increase in PTV<i>V₃</i>_{8.05 Gy}and reduced organs at risk (OAR) doses.<i>Significance.</i>This study demonstrated the feasibility of a CBCT-based automated workflow for breast cancer ART that combines DL-driven dose prediction with automated planning. The approach showed the benefit of replanning each fraction, improved PTV coverage, and offered potential OAR sparing, while also highlighting that daily adaptation may not benefit all patients on all fractions. With appropriately defined thresholds, the workflow can function as a decision-support tool, providing a practical framework to guide adaptive decision-making and optimize the use of clinical resources.