Fetal Head Detection in 2D Ultrasound Images using Deep Learning

Abstract

Automatic fetus head’s boundary detection in the ultrasound image is a very challenging task due to the formation of intensity profile of the image. Ultrasound images are mostly affected by speckle noise and does not show a smooth intensity variation. A computer-based assistance system for the calculation of gestational age of the fetus is useful for the practitioners. Gestational age is a key quantity for an analysis about the baby’s healthy growth which is calculated from the fetus head circumference. In this study, a deep learning-based solution is proposed where the layers and the parameters are adjusted in the U-Net architecture to maximize accuracy of the localization of the head region. There after the extracted contour is used to fit an ellipse and measure the age. The algorithm is trained with 899 images and validated with 100 images. Testing results on 335 images reveals promising with almost 100% localization accuracy and 88.96% specification. Detection of small size fetus is affected by the very closest similar intensity pattern. It will be addressed in the future