OBJECTIVE: Radio frequency ablation (RFA) can be used to treat liver cancer minimally invasively by depositing energy from the RF probe placed in the center of the tumor. The procedure relies on pre-operative imaging (typically MRI or CT) for the interventional planning and ultrasound (US) for intra-operative guidance during needle insertion. Visual presentation of co-registered pre- and intra-operative images would help to improve the navigation during the needle positioning phase. METHODS: In the present study, we compared six registration methods using different similarity metrics: two versions of the correlation ratio, bivariate correlation ratio, and conventional normalized mutual information and correlation coefficient. The accuracy, robustness and speed were assessed by computing rigid registrations between eight pairs of the MR and freehand 3D US datasets. RESULTS: The correlation ratio computed on the MR-gradient-norm and US images outperformed other similarity metrics in terms of robustness (40-82%) and demonstrated average accuracy (0.32 degrees , 0.69 mm) which is clinically acceptable for the RFA of liver cancer. CONCLUSIONS: We observed that the performance of all similarity metrics is largely dependent on the quality of the US images, sufficient field of view of the reconstructed 3D US and absence of motion artifacts.
OBJECTIVE: Radio frequency ablation (RFA) can be used to treat liver cancer minimally invasively by depositing energy from the RF probe placed in the center of the tumor. The procedure relies on pre-operative imaging (typically MRI or CT) for the interventional planning and ultrasound (US) for intra-operative guidance during needle insertion. Visual presentation of co-registered pre- and intra-operative images would help to improve the navigation during the needle positioning phase. METHODS: In the present study, we compared six registration methods using different similarity metrics: two versions of the correlation ratio, bivariate correlation ratio, and conventional normalized mutual information and correlation coefficient. The accuracy, robustness and speed were assessed by computing rigid registrations between eight pairs of the MR and freehand 3D US datasets. RESULTS: The correlation ratio computed on the MR-gradient-norm and US images outperformed other similarity metrics in terms of robustness (40-82%) and demonstrated average accuracy (0.32 degrees , 0.69 mm) which is clinically acceptable for the RFA of liver cancer. CONCLUSIONS: We observed that the performance of all similarity metrics is largely dependent on the quality of the US images, sufficient field of view of the reconstructed 3D US and absence of motion artifacts.