SEMANTIC CONSTRAINED BIDIRECTIONAL TIME SERIES DENOISING ALGORITHM FOR TRUS IMAGE SEGMENTATION

In order to realize the fast location and segmentation of prostate ultrasound image, a segmentation algorithm based on bidirectional time series denoising under semantic constraints is proposed. The shape space was constructed by the point distribution model (PDM) and principal component analysis (PCA) to obtain the centralized shape, and the centralized shape was located by the transformation location matrix obtained by the location convolution network. The shape after localization was expressed by semantic constraint matrix. The cost function and bidirectional time series denoising algorithm were combined to obtain the final segmented image. The experimental results show that this method has better segmentation performance than deep learning algorithms such as Unet, DeepLabV3+, and its average Dice similarity coefficient (DSC) is 0.9679. This method achieves a good balance between the real boundary and the noisy area, reduces the cost and guarantees the accuracy and speed of segmentation.