In cancer treatment by means of radiation therapy having an accurate estimation of tumour size is vital. At present, the tumour shape and boundaries are defined manually by an oncologist as this cannot be achieved using automatic image segmentation techniques. Manual contouring is tedious and not reproducible, e.g. different oncologists do not identify exactly the same tumour shape for the same patient. Although the tumour changes shape during the treatment due to effect of radiotherapy (RT) or progression of the cancer, follow up treatments are all based on the first gross tumour volume (GTV) shape of the tumour delineated before treatment started. Re-contouring at each stage of RT is more complicated due to less image information being available and less time for re-contouring by the oncologist. The absence of gold standards for these images makes it a particularly challenging problem to find the best parameters for any segmentation model. In this paper a level set model is designed for the follow up RT image segmentation. In this contribution instead of re-initializing the same model for level sets in vector-image or multi-phase applications, a combination of the two best performing models or the same model with different sets of parameters can result in better performance with less reliance on specific parameter settings.