The development of magnetic resonance (MR) imaging systems has been extended for the entire radiotherapy process. However, MR images provide voxel values that are not directly related to electron densities, thus MR images cannot be used directly for dose calculation. The aim of this study is to investigate the feasibility of dose calculations to be performed on MR images and evaluate the necessity of re-planning.

A prostate cancer patient was imaged using both MR and computed tomography (CT). The multilevel threshold (MLT) algorithm was used to categorise voxel values in the MR images into three segments (air, water and bone) with homogeneous Hounsfield units (HU). An intensity-modulated radiation therapy plan was generated from CT images of the patient. The plan was then copied to the segmented MR datasets and the doses were recalculated using pencil beam (PB) and collapsed cone (CC) algorithms and Monte Carlo (MC) modelling.

γ Evaluation showed that the percentage of points in regions of interest with γ<1 (3%/3 mm) were more than 94% in the segmented MR. Compared with the planning CT plan, the segmented MR plan resulted in a dose difference of –0·3, 0·8 and –1·3% when using PB, CC and MC algorithms, respectively.

The segmentation and conversion of MR images into HU data using the MLT algorithm, used in this feasibility study, can be used for dose calculation. This method can be used as a dosimetric assessment tool and can be easily implemented in the clinic.

Cone beam computed tomography (CBCT) images contain more scatter than a conventional computed tomography (CT) image and therefore provide inaccurate Hounsfield units (HUs). Consequently, CBCT images cannot be used directly for dose calculation. The aim of this study is to enable dose calculations to be performed with the use of CBCT images taken during radiotherapy and potentially avoid the necessity of re-planning.

A phantom and prostate cancer patient with a metallic prosthetic hip replacement were imaged using both CT and CBCT. The multilevel threshold algorithm was used to categorise pixel values in the CBCT images into segments of homogeneous HU. The variation in HU with position in the CBCT images was taken into consideration and the benefit of using a larger number of materials than typically used in previous work has been explored. This segmentation method relies upon the operator dividing the CBCT data into a set of volumes where the variation in the relationship between pixel values and HUs is small. A field-in-field treatment plan was generated from the CT of the phantom. An intensity-modulated radiation therapy plan was generated from CT images of the patient. These plans were then copied to the segmented CBCT datasets with identical settings and the doses were recalculated and compared.

In the phantom study, γ evaluation showed that the percentage of points falling in planning target volume, rectum and bladder with γ<1 (3%/3 mm) was 100%. In the patient study, increasing the number of bins to define the material type from seven materials to eight materials required 50% more operator time to improve the accuracy by 0·01% using pencil beam and collapsed cone and 0·05% when using Monte Carlo algorithms.

The segmentation of CBCT images using the method in this study can be used for dose calculation. For a simple phantom, 2 values of HU were needed to improve dose calculation accuracy. In challenging circumstances such as that of a prostate patient with hip prosthesis, 5 values of HU were found to be needed, giving a reasonable balance between dose accuracy and operator time.