RATIONALE AND OBJECTIVES: Interventional magnetic resonance imaging (iMRI) allows real-time guidance and optimization of radiofrequency ablation of pathologic tissue. For many tissues, resulting lesions have a characteristic two-boundary appearance featuring an inner region and an outer hyper-intense margin in both T2 and contrast-enhanced (CE) T1-weighted MR images. We created a geometric model-based semiautomatic method to aid in real-time lesion segmentation, cross-sectional/three-dimensional visualization, and intra/posttreatment evaluation. MATERIALS AND METHODS: Our method relies on a 12-parameter, 3-dimensional, globally deformable model with quadric surfaces that describe both lesion boundaries. We present an energy minimization approach to quickly and semiautomatically fit the model to a gray-scale MR image volume. We applied the method to in vivo lesions (n = 10) in a rabbit thigh model, using T2 and CE T1-weighted MR images, and compared the results with manually segmented boundaries. RESULTS: For all lesions, the median error was < or =1.21 mm for the inner region and < or =1.00 mm for the outer hyper-intense region, values that favorably compare to a voxel width of 0.7 mm and distances between the borders manually segmented by the two operators. CONCLUSION: Our method provides a precise, semiautomatic approximation of lesion shape for ellipsoidal lesions. Further, the method has clinical applications in lesion visualization, volume estimation, and treatment evaluation.
RATIONALE AND OBJECTIVES: Interventional magnetic resonance imaging (iMRI) allows real-time guidance and optimization of radiofrequency ablation of pathologic tissue. For many tissues, resulting lesions have a characteristic two-boundary appearance featuring an inner region and an outer hyper-intense margin in both T2 and contrast-enhanced (CE) T1-weighted MR images. We created a geometric model-based semiautomatic method to aid in real-time lesion segmentation, cross-sectional/three-dimensional visualization, and intra/posttreatment evaluation. MATERIALS AND METHODS: Our method relies on a 12-parameter, 3-dimensional, globally deformable model with quadric surfaces that describe both lesion boundaries. We present an energy minimization approach to quickly and semiautomatically fit the model to a gray-scale MR image volume. We applied the method to in vivo lesions (n = 10) in a rabbit thigh model, using T2 and CE T1-weighted MR images, and compared the results with manually segmented boundaries. RESULTS: For all lesions, the median error was < or =1.21 mm for the inner region and < or =1.00 mm for the outer hyper-intense region, values that favorably compare to a voxel width of 0.7 mm and distances between the borders manually segmented by the two operators. CONCLUSION: Our method provides a precise, semiautomatic approximation of lesion shape for ellipsoidal lesions. Further, the method has clinical applications in lesion visualization, volume estimation, and treatment evaluation.