This project aims to increase the efficiency of the IGBT treatment planning process by determining a suitable technique for the auto segmentation of ROIs. The use of an auto-segmentation solution also aims to increase reproducibility in ROI boundary definition and consequently increase accuracy in dose reporting (Valentini et al. 2014). The current clinical IGBT treatment planning process typically takes a couple of hours with one of the more time consuming components being the outlining of ROIs on the MRI images (Valentini et al. 2014; Sharp et al. 2014). This comprises the delineation of the target volumes by the oncologist and the delineation of the organs at risk (OARs) by the physics operator. The outlining of ROIs can typically take 40-60 minutes depending on patient anatomy and image quality. Outlining of ROIs is susceptible to high operator variability (Valentini et al. 2014; Sharp et al. 2014).During treatment planning the highest dose being received by 2 cubic centimetres of an OAR (D2cc) is used in the optimisation of the plan to ensure it is lower than the specified tolerance. Any user variation in OAR outlining can therefore result in differences in the treatment plan delivered to the patient. Accurate delineation of OARs as well as target volumes is therefore imperative to successful radiotherapy (Fritscher et al. 2014). Although there is some evidence of auto-segmentation of ROIs for brachytherapy this currently appears to be limited to prostate treatments (Martin et al. 2008; Nouranian et al. 2015). The first proposed step of the project is to perform an audit of the current time take to segment ROIs and analyse inter-operator variability. At Velindre Cancer Centre we have two oncologists performing target volume delineation and nine physics operators delineating OARs. The variation between individual oncologists will be analysed and the physics operator OAR contours will be compared to an expert user. IGBT with paraxial MRI started at Velindre Cancer Centre in late 2014 with over 160 patients treated since then, equalling over 460 outlined image sets. These outlined MRI image sets will be used to establish a ‘gold standard’ database to inform the development of the auto-segmentation solution. Depending on the findings from the variability study the image sets may need adjusting to ensure consistency of delineation. To develop a clinically acceptable auto segmentation solution a number of possible options will be assessed for their feasibility of application to paraxial MRIs (Fritscher et al. 2014). Firstly we are currently in discussions with a commercial provider to create a working partnership to facilitate the development of an auto-segmentation solution for MRI for IGBT. This particular commercial partner already has an established memorandum of understanding with our Trust. Secondly there are a number of open source software packages that could be investigated, such as ITK-SNAP, 3D Slicer, FIJI and MITK. It is anticipated the literature review will enable an assessment of the different open source options and their feasibilities as alternatives to the commercial solution discussed above. Once a suitable auto-segmentation solution has been developed its suitability for clinical use needs to be evaluated. Appropriate evaluation metrics, such as the DICE similarity coefficient, will be determined from the literature to assess the performance of the developed auto-segmentation solution against manual clinical delineations by an expert operator, one of our IGBT oncologists (Valentini et al. 2014). An analysis of the significance of geometric variations in ROI delineations will be performed by assessing their effect on dose constraints, such as OAR doses and target coverage. The audit of time taken to segment ROIs will be performed again to compare auto-segmentation delineation time to manual segmentation time.