Ensemble Transfer Learning for Fetal Head Analysis: From Segmentation to Gestational Age and Weight Prediction

Abstract

Ultrasound is one of the most commonly used imaging methodologies in obstetrics to monitor the growth of a fetus during the gestation period. Specifically, ultrasound images are routinely utilized to gather fetal information, including body measurements, anatomy structure, fetal movements, and pregnancy complications. Recent developments in artificial intelligence and computer vision provide new methods for the automated analysis of medical images in many domains, including ultrasound images. We present a full end-to-end framework for segmenting, measuring, and estimating fetal gestational age and weight based on two-dimensional ultrasound images of the fetal head. Our segmentation framework is based on the following components: (i) eight segmentation architectures (UNet, UNet Plus, Attention UNet, UNet 3+, TransUNet, FPN, LinkNet, and Deeplabv3) were fine-tuned using lightweight network EffientNetB0, and (ii) a weighted voting method for building an optimized ensemble transfer learning model (ETLM). On top of that, ETLM was used to segment the fetal head and to perform analytic and accurate measurements of circumference and seven other values of the fetal head, which we incorporated into a multiple regression model for predicting the week of gestational age and the estimated fetal weight (EFW). We finally validated the regression model by comparing our result with expert physician and longitudinal references. We evaluated the performance of our framework on the public domain dataset HC18: we obtained 98.53% mean intersection over union (mIoU) as the segmentation accuracy, overcoming the state-of-the-art methods; as measurement accuracy, we obtained a 1.87 mm mean absolute difference (MAD). Finally we obtained a 0.03% mean square error (MSE) in predicting the week of gestational age and 0.05% MSE in predicting EFW.

Evolving the Era of 5D Ultrasound? A Systematic Literature Review on the Applications for Artificial Intelligence Ultrasound Imaging in Obstetrics and Gynecology

“…This literature review includes 10 studies on the assessment of GA, starting in 1996 with a pioneering study of Beksaç et al on the estimation of GA via the calculation of the fetal biparietal diameter and HC [146]. In addition to this, the studies of Namburete et al and Alzubaidi et al similarly used the anatomy and growth of the fetal head for GA estimation [147,148]. Dan et al developed a DeepGA model that used the three main factors of fetal head, abdomen, and femur [149], while Lee et al proposed a machine learning method to accurately estimate GA with standard US planes [150].…”

“…The purpose of these studies, which were based on US data from the US and Zambia, was the successful establishment of an AI algorithm for GA estimation from simplified blind US sweeps of US novices in low-resource countries [144,145,152,153]. The benefits of the GA AI models were the possibility for application in low-resource countries [144,145,148,149,152], even without internet connectivity [145], and in portable devices [148], promising high accuracy with an error of 3.9 to 5 days in GA estimation [149,152]. An important limitation of the AI models was described to be application in very early [144,147] or very late stages of pregnancy [146,147,152], the latter of which was due to the thickened texture of the fetal skull.…”

“…It, therefore, is of public health relevance, by reducing costs and the need for sonography experts [35,39]. In this framework, AI models are additionally able to enhance image acquisition and diagnostic accuracy in point-of-care US devices, which is of particular significance for application in low-resource settings and medically underserved areas [34,145,148,199].…”

Enhancing Fetal Anomaly Detection in Ultrasonography Images: A Review of Machine Learning-Based Approaches

“…Other valuable information can be drawn from the segmentation of fetal head images in obstetrics for monitoring fetal growth [132]. This information is valuable for the assessment of fetal health.…”

Multimodal Early Birth Weight Prediction Using Multiple Kernel Learning

“…These algorithms were fed with variables obtained before delivery, among which anthropometric data of the mother, mother’s education and systolic and diastolic blood pressure in the second and third trimesters of gestation stand out. Alzubaidi et al [ 12 ] proposed a methodology for segmentation and extraction of indicators from US images of the fetal head using neural networks. They trained several regression models (RF, Linear Regression, SVR, LASSO, Voting Regressor or Deep NN among the main ones), using metrics such as Biparietal Diameter (BPD), Head Circumference (HC) and Occipitofrontal Diameter (OFD) to estimate the fetal weight at the time of US acquisition.…”