Classification of common fetal anatomical planes from ultrasound imaging using dempster shafer theory and deep learning

Abstract

Ultrasound (US) examination is a widely used important instrument to monitor mother and fetus health in a cost-effective and non-invasive way. The acquisition of Ultrasound (US) images to determine vital fetal organs for the screening of fetal abnormalities requires identifying the exact plane and region of the desired organs. Even after following guidelines from appropriate committees, a sonologist some- times may have difficulties in acquiring an excellent fetal plane image or make errors in judgement for several reasons like inexperienced operators, faulty equipment or movement of the fetus. Furthermore, sometimes due to the fetus being in critical positions or due to the increase of adipose tissue inside the mother, it can create various problems in the imaging like artifacts, acoustic shadows or even low signal to noise ratio. Also, in an appropriate institute, a specialist of fetal images reviews the sonographer’s analysis and chooses images that contains structures of interest which later gets reviewed by a senior maternal-fetal expert or a specialist doctor. This is a manual process which is expensive, cumbersome and sensitive to mistakes. So we propose a method that combines Convolutional Neural Network (CNN) and Dempster-Shafer theory (DST) to create a DST based evidential classifier or evidential CNN called E-CNN for the classification of common fetal anatomical planes like brain, abdomen, thorax, femur as well as the maternal cervix from its ultrasound images.