AUTOMATIC DETECTION OF LOCAL FETAL BRAIN STRUCTURES IN ULTRASOUND IMAGES

Although ultrasound imaging is the main modality to study fetal growth, the analysis of such images is an under-studied area of research especially using 3D ultrasound. In this paper we propose an automatic technique to locate four local fetal brain structures in 3D ultrasound images. Clinically, the localization of such structures in 3D is hard and subjective. However, structure localization is critical to detect some brain abnormalities and to do fetal biometry. The technique we propose is based on a discriminative model (Random Forests) which is gaining a lot of interest recently. The novelty of this work lies in 1) the implicit integration of image features and the relative position of structures in the fetal brain within the technique and 2) the application itself since it is very challenging. We report promising first results which are consistent with published literature on visual detection of fetal brain structures, and suggest that automated analysis of 3D fetal neurosonography may be possible in the future.