Pseudo-CT scan creation using registration of transabdominal ultrasound volumes of a prostate cancer patient

Purpose: A crucial step in radiotherapy workflows is patient setup. The patients’ position prior to each treatment fraction (TX) must be as similar as possible to the simulation CT scan (CTsim), as the treatment plan was based on this scan. Even when the patient seems correctly aligned externally, internal tissue distributions might have changed. Delivery of the initial treatment plan might then result in suboptimal dose depositions. As frequent CT scan (CTtx) acquisition for treatment plan verification is not feasible, we propose to create instead pseudo-CT scans using simulation ultrasound (USsim) to treatment ultrasound (UStx) registration.

Methods: Ten data pairs USsim/CTsim - UStx/CTtx were considered using five co-registered CT and transabdominal ultrasound data sets from a prostate cancer patient. Eight different approaches, combinations of rigid translation and/or deformable registration, were used to create deformation fields between USsim and UStx for each data pair. These fields could then be applied to the corresponding CTsim, resulting in the creation of a pseudo-CT scan. The pseudo-CT scans were compared anatomically to the CTsim, to assess which quantitatively represented the anatomy of the patient at TX more reliably.

Results: In five out of ten data pairs, all eight approaches resulted in the creation of a pseudo-CT volume equally or more similar to CTtx than CTsim, with an average improvement of 54.1% in Dice similarity coefficient (DSC) and 35.7% in gamma index based on gray values. For the remaining pairs, four up to seven approaches resulted in an improvement in both measures. Overall, the DSC improvements varied between 4.3% and 126.5%, while the gamma index improvements varied between 1.1% and 58.4%.

Conclusion: For all studied data pairs, at least four out of eight explored approaches resulted in a pseudo-CT volume which was more representative of the patient at TX than the CTsim volume.