Two-Stage Machine Learning Pipeline for Fetal Head Analysis

Abstract

Background: The accurate measurement of fetal head circumference (HC) and the precise estimation of gestational age are vital components of prenatal assessment. However, manual techniques are often labor-intensive and subject to variability among different observers. Automated methods have the potential to enhance measurement consistency and reduce the clinical workload. Purpose: This study presents an integrated two-stage machine learning pipeline aimed at automating fetal head segmentation and trimester-specific gestational age classification using ultrasound data. Methods: The proposed framework is composed of two stages. In the first stage, fetal head segmentation is conducted using a Random Forest classifier that leverages handcrafted pixel-level features. In the second stage, features related to shape, intensity, and texture are extracted from the segmented region and classified using a top-3 ensemble of LightGBM, Support Vector Machine with a Radial Basis Function (RBF) kernel, and XGBoost, with adjustments for class imbalance and feature optimization. Results: The segmentation framework achieved a mean accuracy of 89.02%. The trimester classification model achieved an overall accuracy of 85.62%, a precision of 86.57%, a recall of 85.62%, and an F1-score of 85.59%. Conclusion: The proposed two-stage pipeline effectively achieves precise segmentation of the fetal head and accurate trimester classification, highlighting its considerable potential for integration into clinical decision-support systems, especially in healthcare environments with constrained resources.