CardioNet : Identification of fetal cardiac standard planes from 2 D Ultrasound data

Congenital Heart Defects (CHDs) affect up to 1% of newborns globally and about 4-5 % in developing world. Identifying the fetal cardiac standard planes from 2D ultrasound videos is the preliminary is the first step towards the diagnosis of CHDs. The task of manually identifying the standard cardiac planes such as four Chamber, three vessel, RVOT, LVOT from video frames requires extensive training and experience. In India, about 69 % of the entire population reside in rural areas & the acute shortage of trained and experienced sonographers in these parts of the country make detection of CHDs extremely difficult. In this paper, we propose a novel deep learning based approach for identification of fetal cardiac standard planes from 2D ultrasound videos. Such a tool would help not only the novice sonographers but also the experts in identifying all the cardiac standard planes from a free hand fetal ultrasound video. The dataset was acquired from a collaborating institute post appropriate ethical clearance. On the test data (n= 1636 slices), for the task of identifying the four standard planes, the proposed network achieved an accuracy of 92 % and an average F1-score of 0.919.