Automatic prostate segmentation in MR images based on 3D active contours with shape constraints

Planning radiotherapy of prostate cancer requires the prostate segmentation in computed tomography (CT) images that can be manual (done by medical doctors), semi-automatic or automatic. Additional usage of magnetic resonance (MR) images, where the soft tissue are better visible, makes this operation more robust. The paper addresses the problem of prostate segmentation in MR data. Its main contribution relies on novel application of the well-known active contour (AC) method with gradient vector flow (GVF) modification to this task. It is shown in the paper that such approach is successful only after addition of a priori knowledge in the form of prostate shape constraint. The statistical prostate shape was modeled as an ellipse which parameters are calculated exploiting statistical atlas principles. It is presented using Dice similarity measure that the proposed automatic prostate segmentation offers results that are very close to the manual ones and can be used in radiotherapy planning.