A framework for analysis of linear ultrasound videos to detect fetal presentation and heartbeat.

M.A Maraci,A Papageorghiou,C.P Bridge,R Napolitano,J.A Noble

doi:10.1016/j.media.2017.01.003

M.A Maraci, A Papageorghiou + Show 3 more

Open Access

https://doi.org/10.1016/j.media.2017.01.003

Copy DOI

Abstract

Confirmation of pregnancy viability (presence of fetal cardiac activity) and diagnosis of fetal presentation (head or buttock in the maternal pelvis) are the first essential components of ultrasound assessment in obstetrics. The former is useful in assessing the presence of an on-going pregnancy and the latter is essential for labour management. We propose an automated framework for detection of fetal presentation and heartbeat from a predefined free-hand ultrasound sweep of the maternal abdomen. Our method exploits the presence of key anatomical sonographic image patterns in carefully designed scanning protocols to develop, for the first time, an automated framework allowing novice sonographers to detect fetal breech presentation and heartbeat from an ultrasound sweep. The framework consists of a classification regime for a frame by frame categorization of each 2D slice of the video. The classification scores are then regularized through a conditional random field model, taking into account the temporal relationship between the video frames. Subsequently, if consecutive frames of the fetal heart are detected, a kernelized linear dynamical model is used to identify whether a heartbeat can be detected in the sequence. In a dataset of 323 predefined free-hand videos, covering the mother's abdomen in a straight sweep, the fetal skull, abdomen, and heart were detected with a mean classification accuracy of 83.4%. Furthermore, for the detection of the heartbeat an overall classification accuracy of 93.1% was achieved.

Highlights

There have been significant advances in the analysis of ultrasound images in the last decade due in part to increased image quality and the introduction of modern machine learning into the medical image analysis field (Noble, 2016)
In previous works on which the current paper builds, we have investigated the bag of visual words approach with feature symmetry filters (Maraci et al, 2014a) as well as improved Fisher vector (IFV) encoding (Maraci et al, 2015) with a support vector machine (SVM) to identify frames of interest in an ultrasound video
The first experiment evaluated the accuracy of the frame classification task, including the use of different lowlevel features and SVM kernels

Summary

Introduction

There have been significant advances in the analysis of ultrasound images in the last decade due in part to increased image quality and the introduction of modern machine learning into the medical image analysis field (Noble, 2016). Machine learning is arguably very well-suited to recognize sonographic patterns in ultrasound images, which can form the basis of image-based decision-making. Traditional biomedical image analysis methods can find the dropouts, shadows, and sonographic signatures characteristic of ultrasound images difficult to accommodate, as they are the mapping of anatomy through the ultrasound image formation process. The most successful traditional methods in the literature are model-based methods that use strong geometric models as priors to cope with missing boundaries and artefacts. The majority of the image analysis literature in this area has focused on automation of fetal biometry measurement for the anomaly scan

Objectives

Methods

Results