Magnetic resonance imaging and transrectal ultrasound prostate image segmentation based on improved level set for robotic prostate biopsy navigation

Aim: Transrectal ultrasound (TRUS) guided prostate biopsy is a typical early prostate examination. However, the ultrasound imaging suffers from blurred contour, intensity inhomogeneity and small surrounding soft tissue differentiation. To take advantage of clear magnetic resonance imaging (MRI) into robotic prostate biopsy navigation, the prostate regions in the MRI and TRUS images need to be segmented separately. This paper proposes an improved level set segmentation model based on prior shape, which aims to provide a better solution to the prostate segmentation problems in TRUS and MRI. Methods: In our segmentation model, the Gaussian probability model is used to establish the statistical learning of the prior shape, and the cosine function is used to represent the energy term fitting of the traditional prior shape and the local intensity information. Results: The experiment results show that the model can adapt to different forms of prostate in MRI and TRUS more accurately, and the prostate biopsy accuracy in our biopsy system can reach 2.24 ± 1.44 mm. Conclusion: This segmentation model has high accuracy, meets the clinical needs in robotic prostate biopsy navigation.