Non-rigid registration of planning and post-ablation CT images of the liver for assessing geometric ablation accuracy and predicting treatment outcome.

Purpose: For percutaneous radio frequency ablation (RFA), accurate placement of the ablation needle to ensure complete coverage of the lesion by the ablative zone, is considered the primary indicator for treatment success. We propose a method for early assessment of the geometrical accuracy of the treatment using CT images acquired before and within 24h after ablation, and evaluate the predictive value of this measure for the treatment outcome.

Materials and Methods: For all patients, we acquired a whole-body planning PET/CT scan within one month before RFA, a liver-only PET/CT scan within 24 hours post-RFA, and follow-up whole body PET/CT scans every 2-3 months after the intervention - all with intravenous iodized contrast administration. RFA's were performed under CT fluoroscopic guidance. Planning and post-ablation CT images were aligned using commercial registration software (Integrated Registration, GE Healthcare, Milwaukee, WI, USA). Lesion and ablation zone were semi-automatically segmented and masked during registration. A global, non-rigid registration based on mutual information was performed, followed by an interactive, rigid registration based on a smaller region of interest. Using the registered images, we verified the geometrical accuracy of the RFA treatment by measuring the minimal distance between the lesion and the outer edge of the ablation zone, and correlated this to local tumor progression (LTP) as recorded during follow-up.

Results: 23 patients, with a total of 45 liver lesions, were included in the study. A total of 11 liver lesions showed LTP during follow-up imaging. For 5 lesions, the result given by the registration procedure were not fully satisfying, including one for which the registration was deemed to have failed. Based on the aligned images, 29 lesions fell fully within the ablation zone, while 10 did not. For 6 lesions, the edge was found to coincide with the edge of the ablation zone. When considering lesions found at the edge as not fully covered, full coverage of the lesion was found to be a powerful predictor for LTP (Se = 100%, Sp = 85%, PVV = 69%, NPV = 100%) and outperformed predictions based on the PET/CT taken within 24h after RFA to detect residual lesion (McNemar, p
Conclusions: Verifying the geometrical placement of the ablation needle through non-rigid registration of planning and early post-ablation CT images is feasible and has a strong predictive power for treatment outcome. Increasing the robustness and degree of automation of the procedure could further improve accuracy and reproducibility of the method.

Clinical Relevance: Early accurate detection of treatment failure will allow for reablation and will ultimately improve the prognosis of the minimally invasive procedure.