Predicting FGR at Early Pregnancy via a HDC-based Encoding Method

Fetal growth restriction (FGR) is a leading cause of fetal complications and perinatal mortality. Current research primarily focuses on diagnosing FGR through ultrasound-based fetal weight estimation and blood flow measurements. However, these methods are limited in detecting FGR in the early stages of pregnancy. Studies have shown functional differences between normal and FGR placentas; however, these differences are subtle in ultrasound images. To fully explore the relationship between placental features and FGR, and to better extract and amplify the distinctions between FGR cases and normal ones, this paper proposes a novel ultrasound image encoding method inspired by hyperdimensional computing (HDC) and the Bag of Visual Words (BOVW). It encodes ultrasound regions using an HDC-based approach, which constructs a visual word bag in a high-dimensional space and then utilizes them for early FGR detection. Experimental results based on a dataset approved by the ethics committee demonstrate that the effectiveness of the method in early FGR detection achieves an accuracy of 74.3%, surpassing existing state-of-the-art models.