Contour-based Medical Image Fusion for Biopsy
Hongyue Wu, Yunjie Chen, Biao Huang, Su Zhang, Yunkai Zhu, Yaqing Chen · 2020
Implementing prostate biopsy with the guidance of magnetic resonance imaging (MRI) can reduce the risk and improve the accuracy in prostate cancer therapy. How to combine information from MRI and transrectal ultrasonography (TRUS) plays an important role in image fusion-guided biopsy. In this study, we introduce a neural network based on U-net for prostate segmentation in MRI and TRUS. On the basis of the contour generated from segmentation masks, we implement the thin plate spline-robust point matching for non-rigid registration. We validate our method using animal experiments with simulated lesions and obtain convincing results. The results show that our model achieve a mean intersection over union (mIOU) of 0.819 for TRUS segmentation and a mIOU of 0.878 for MRI segmentation. We also modify root mean square error (RMSE) and achieve the result of 0.0082 in registration which is more accurate than rigid registration.