|Title||Various approaches for pseudo-CT scan creation based on ultrasound to ultrasound deformable image registration between different treatment time points for radiotherapy treatment plan adaptation in prostate cancer patients|
|Publication Type||Journal Article|
|Year of Publication||2016|
|Authors||Camps, S, van der Meer, S, Verhaegen, F, Fontanarosa, D|
|Journal||Biomedical Physics and Engineering Express|
The purpose of this study was to evaluate eight possible approaches to create pseudo-CT images for radiotherapy (RT) treatment re-planning. These re-planning CT scans would normally require a separate CT scan session. If important changes occur in patient’s anatomy between simulation (SIM) and treatment (TX) stages, 3D ultrasound (US) images acquired at the two stages, available in US guided RT workflows, can be used to produce a deformation field. Proof of concept research showed that the application of this deformation field to the SIM CT image yields a pseudo-CT which can be more representative of the patient at TX than SIM CT. Co-registered CT and US volumes acquired at five different time points during the RT course of a prostate cancer patient were combined into data pairs, providing ground truth CT images (CTtx). Eight different methods were explored to create the deformation field that was used to produce the pseudo-CT scan. Anatomical structure comparison and γ index calculations were used to compare the similarity of the pseudo-CT volumes and the reference TX CT volumes.
In five out of ten data pairs, all the eight approaches resulted in the creation of a pseudo-CT equally or more similar to the TX CT than the SIM CT within the region of interest, with an average improvement of 54.1% (range: 5.1% – 126.5%) in dice similarity coefficient (DSC) and 32.3% (range: 0.3% – 52.6%) in γ index. For the remaining data pairs, four up to seven approaches resulted in an improvement in both DSC (range: 4.3% - 54%) and γ index (range: 0.8% - 41.3%).
In conclusion, at least four out of eight explored approaches resulted in more representative pseudo-CT images in all the data pairs. In particular, the approaches in which an initial rigid alignment was combined with deformable registration performed best.