A hierarchical model for automatic nuchal translucency detection from ultrasound images

Abstract

The nuchal translucency (NT) thickness is an important parameter in the diagnosis of fetuses. The previous computerized methods often require manual operations to select the NT region, which leads to the time-consuming problem and the detection variability. In the paper, a hierarchical model is proposed for the automated detection of the NT region. Three discriminative classifiers are first trained with Gaussian pyramids to represent the NT, head and body of fetuses respectively. Then a spatial model is proposed to denote the spatial constrains among them. Finally the dynamic programming and generalized distance transform are applied for the inference from the proposed model, which ensures that the optimal solution can be obtained for the NT detection. The direction problem of fetuses is resolved by the introduced "OR" node. The performance of the proposed model is verified by the experimental results of 690 clinical NT ultrasound images.