Development of Implicit Representation Method for Freehand 3D Ultrasound Image Reconstruction of Carotid Vessel

The objective of this study is to develop a novel 3D ultrasound reconstruction algorithm based on deep learning method, which can reduce the artifacts of 3D volume of carotid artery. The method implicitly inferred and represented the intensity and semantic probability of the voxel in the image volume using a multi-layer perceptron (MLP) network. Fifteen datasets were collected on the carotid atherosclerosis (CA) and healthy subjects in clinic and reconstructed to 3D volume after vessel segmentation. The discontinuity and distortion were defined to evaluate the artifact of the images volume, and the variation of overall curvature was calculated to evaluate the smoothness of the reconstructed vessel wall. The results showed our method outperformed traditional methods by reducing 34.79% distortion and 61.61 % discontinuity and yielding more than 60% improved cases. It demonstrated that our method has the potential of providing smoother and more continuous image volume comparing with the conventional reconstruction method.