Abstract Identifying the appropriate clinical target volume (CTV) to capture microscopic disease is the greatest limitation in clinical radiotherapy in efforts to offer maximally conformal treatment to minimize radiation associated toxicity. The challenge of defining the CTV comes from inherent uncertainty in the tumor spread beyond the visible gross tumor volume (GTV). Delineation of the CTV is a laborious manual process. Furthermore, there exists a practical disconnect between CTV contouring and the subsequent treatment plan dose optimization. Exploration of the real tradeoff between covering malignancy with the dose effective for tumor control and delivering potentially toxic dose to surrounding healthy tissues is currently impossible. The broad long-term goal of this project is to make CTV definition easier and better. We will focus on two challenging disease sites, glioma and sarcoma. Our methods can be generalized to essentially all other disease sites. The first aim is to automate CTV definition. This will be accomplished by machine learning of barrier structures and anatomic domains that are known to affect the spread of tumor beyond the visible GTV. The CTV will be expanded in 3D taking the preferred direction of spread in the different anatomic domains (such as spread along muscle fibers) into account. The second aim is to develop a user interface that lets the user interact with the automatic CTV definition system, to avoid a black box impression. The user can edit the auto generated contour if necessary. Any changes will be logged and used to retrain the system. The CTV expansion will be integrated in a multi- criteria optimization system for treatment planning, where the user can interactively explore the dosimetric impact of CTV expansion on the dose coverage of the tumor and dose burden in normal structures. In the third aim we will test the hypotheses that this system will lead to a more consistent definition of the CTV, better time efficiency, and better treatment plans leading to provable improvements of the expected clinical outcome. We will make the system available as a standalone system to academic users and hospitals.