Robust Fetal Head Circumference Calculation on Ultrasound Images Using Just-in-Time Compiled Ellipse Fitting

Calculation of fetal head circumference (HC) is essential for various purposes, particularly in evaluating fetal health during prenatal care. However, performing HC calculations manually is labor intensive, prone to errors, and requires specific expertise. Thus, there is a pressing demand for automated methods for HC calculation. Recent automation methods for head circumference calculation focus on using deep learning that is often slow and memory hog due to the high computational complexity. This study presents a novel approach for fast and efficient automatic HC calculation from 2D ultrasound images by leveraging just-in-time compiled ellipse fitting based on Randomized Hough Transform. Before ellipse fitting, there are downscaling, blurring, and clean edge detection to increase robustness. After ellipse fitting, there is circumference calculation using the Ramanujan approximation. Our approach achieved a remarkable speed of 0.41 s per image with less than 800 kB of memory. It demonstrated robust accuracy with a Hausdorff distance of 3.88 mm, a Dice score of 0.93, and an absolute difference of 7.24 mm, which is comparable to existing methods.