A 3D lesion model for segmentation and evaluation of interventional MRI-guided thermal ablation treatment

We are investigating imaging considerations and physiologic response to radiofrequency ablation of pathologic tissue, where the lesion is localized and monitored through realtime MR images. For this, we created a twelve parameter deformable model of lesion shape. The model is useful for lesion visualization, volume approximation, automated segmentation, and volume registration. Using digital phantoms we quantified the model's ability to estimate whole lesion shape from erroneous and partial boundary information. Among all in vivo lesions (n=10), the mean median fit errors were 0.32 mm (0.11 SD) and 0.38 mm (0.16 SD) for the inner and outer surfaces respectively.