Application of artificial neural networks to evaluate femur development in the human fetus

Abstract

The present article concentrates on an innovative analysis that was performed to assess the development of the femur in human fetuses using artificial intelligence. As a prerequisite, linear dimensions, cross-sectional surface areas and volumes of the femoral shaft primary ossification center in 47 human fetuses aged 17–30 weeks, originating from spontaneous miscarriages and preterm deliveries, were evaluated with the use of advanced imaging techniques such as computed tomography and digital image analysis. In order to ensure the data representativeness and to avoid introducing any hidden structures that may exist in the data, the entire dataset was randomized and separated into three subsets: training (50% of cases), testing (25% of cases), and validation (25% of cases). Based on the collected numerical data, an artificial neural network was devised, trained, and subject to testing in order to synchronously estimate five parameters of the femoral shaft primary ossification center, thus leveraging fundamental information such as gestational age and femur length. The findings reveal the formulated multi-layer perceptron model denoted as MLP 2-3-2-5 to exhibit robust predictive efficacy, as evidenced by the linear correlation coefficient between actual values and network outputs: R = 0.955 for the training dataset, R = 0.942 for validation, and R = 0.953 for the testing dataset. The authors have cogently demonstrated that the use of an artificial neural network to assess the growing femur in the human fetus may be a valuable tool in prenatal tests, enabling medical doctors to quickly and precisely assess the development of the fetal femur and detect potential anatomical abnormalities.