Expert trained AI-models as a paradigm shift for disseminating radiotherapy contouring guidelines: The EPTN consensus atlas for CT- and MR-based neuro-oncology contouring

We present an expert-trained deep learning model for automated delineation of 25 organs-at-risk defined in the EPTN atlas for neuro-oncological radiotherapy. Multiple input configurations were evaluated. The dual-input channel model combining contrast-enhanced T1-weighted MRI and CT achieved the best overall results (median surface Dice Similarity Coefficient = 0.84). It demonstrated robust performance across a wide range of organs-at-risk, supporting its use as a segmentation-support tool under human supervision and for fully automated retrospective dose–response analyses. We release the model publicly to the international community, while explicitly acknowledging that clinical use remains the responsibility of each institution within its regulatory context.