Shapes Similarity and Feature Reconstruction Comparison Based Active Contour Model

The medical Internet of things is the foundation of smart medical. Medical image is the main resource for transmission on the medical Internet of things. Ultrasound image, as a primary medical image, is widely used in computer-aided therapy. The segmentation of lesion region in ultrasound image sequences plays a crucial role in computer-aided therapy. Active contour models are widely used in ultrasound image segmentation to extract the lesion boundary through the low level appearance cues of lesion region. However, due to diseases and imaging artifacts, the low level appearance cues might cause weak or misleading features which corrupts the performance of active contour. In this situation, the shape prior becomes a powerful tool to aid active contour to resist the interference with misleading features. However, the various ways to model the prior of shapes are usually learnt from a large set of annotated data, which is not always feasible in practice. It is doubted that the existing shapes in the training set will be sufficient to model the new instance in the testing image. In this paper, a novel active contour based on shape similarity and feature reconstruction comparison is proposed to segmenting ultrasonic image sequence. In our works, the similarity of object shapes in the image sequence is modeled as a shape prior in a active contour model, which can be interpreted as an unsupervised approach of shape prior modeling without a large number of annotated data. Furthermore, a novel sparse representation based object boundary searching strategy, named feature reconstruction comparison, is proposed by exploiting both the low level appearance cues comparison of the object and background to reduce the error of searching, which is also used to resist the defects of ultrasound image. In order to verify the performance of our method, the clinical image sequences were used as the training and test set to validate our method. The proposed method was compared with three well-known methods in the same test set. The results demonstrates that the proposed method can consistently improve the performance of active contour models and increase the robustness against image defects consequently, it improves the efficiency and effect of the computer assisted therapy