Automatic model-based evaluation of magnetic resonance-guided radio frequency ablation lesions with histological correlation

Purpose: To develop a model-based method for automatic evaluation of radio frequency (RF) ablation treatment using magnetic resonance (MR) images.

Materials and methods: RF current lesions were generated in a rabbit thigh model using MR imaging (MRI) guidance. We created a 12-parameter, three-dimensional, globally deformable model with quadric surfaces that delineates lesion boundaries and is automatically fitted to MR grayscale data. We applied this method to in vivo T2- and contrast-enhanced (CE) T1-weighted MR images acquired immediately post-ablation and four days later. We then compared results to manually segmented MR and three-dimensional registered corresponding histological boundaries of cellular damage.

Results: Resulting lesions featured a two-boundary appearance with an inner region and an outer hyperintense margin on MR images. For automated vs. manual MR boundaries, the mean errors over all specimens were 0.19 +/- 0.51 mm and 0.27 +/- 0.52 mm for the inner surface, and -0.29 +/- 0.40 mm and -0.12 +/- 0.17 mm for the outer surface, for T2- and CE T1-weighted images, respectively. For automated vs. histological boundaries, mean errors over all specimens were 0.07 +/- 0.64 mm and 0.33 +/- 0.71 mm for the inner surface, and -0.27 +/- 0.69 mm and 0.02 +/- 0.43 mm for the outer surface, for T2- and CE T1-weighted images, respectively. All boundary errors compared favorably to MR voxel dimensions, which were 0.7 mm in-plane and 3.0 mm thick.

Conclusion: The method is accurate both in describing MR-apparent boundaries and in predicting histological response and has applications in lesion visualization, volume estimation, and treatment evaluation.