®®®® SIIA Público

Título del libro: Proceedings Of The 21st International Symposium On Medical Information Processing And Analysis, Sipaim 2025
Título del capítulo: Convolutional Neural Network Trained with Heatmaps for Segmentation and Measurement of the Femur in Fetal Ultrasound

Autores UNAM:
FERNANDO ARAMBULA COSIO; JORGE LUIS PEREZ GONZALEZ;
Autores externos:

Idioma:

Año de publicación:
2025
Palabras clave:

Biparietal diameters; Convolutional neural network; Femur length; Fetal ultrasound; Fetal weight; Heatmaps; Length measurement; Manual intervention; Measurements of; Sonographers


Resumen:

© 2025 IEEE.Fetal femur length measurement via ultrasound, alongside abdominal circumference, biparietal diameter, and cranial circumference assessments, are key components of prenatal screening, as they enable the estimation of fetal weight and the evaluation of intrauterine development. Currently, this procedure still heavily relies on manual intervention by expert radiologists or sonographers, who must identify anatomical landmarks to perform the measurements. This dependence on specialized human expertise limits diagnostic accessibility in remote areas far from major urban centers. In this work, we propose a novel convolutional neural network trained on binary masks and heatmaps of femur endpoints. The model achieves robust performance, segmenting the femur with an 88% Dice coefficient, and measuring its length with only a 2.515% error margin. This work is part of a broader machine learning based system for fetal weight estimation, aiming to expand diagnostic capabilities to healthcare centers with limited human resources.


Entidades citadas de la UNAM: