Large-scale annotation dataset for fetal head biometry in ultrasound images

This dataset features a collection of 3832 high-resolution ultrasound images, each with dimensions of 959 x661 pixels, focused on Fetal heads. The images highlight specific anatomical regions: the brain, cavum septum pellucidum (CSP), and lateral ventricles (LV). The dataset was assembled under the Creative Commons Attribution 4.0 International li-cense, using previously anonymized and de-identified images to maintain ethical standards. Each image is complemented by a CSV file detailing pixel size in millimeters (mm). For enhanced compatibility and usability, the dataset is available in 11 universally accepted formats, including Cityscapes, YOLO, CVAT, Datumaro, COCO, TFRecord, PASCAL, LabelMe, Segmentation mask, OpenImage, and ICDAR. This broad range of for-mats ensures adaptability for various computer vision tasks, such as classification, segmentation, and object detection. It is also compatible with multiple medical imaging software and deep learning frameworks. The reliability of the annotations is verified through a two-step validation process in-volving a Senior Attending Physician and a Radiologic Technologist. The Intraclass Correlation Coefficients (ICC) and Jac -card similarity indices (JS) are utilized to quantify inter-rater agreement. The dataset exhibits high annotation reliability, with ICC values averaging at 0.859 and 0.889, and JS values at 0.855 and 0.857 in two iterative rounds of annotation. This dataset is designed to be an invaluable resource for ongoing and future research projects in medical imaging and com-puter vision. It is particularly suited for applications in pre-natal diagnostics, clinical diagnosis, and computer-assisted interventions. Its detailed annotations, broad compatibility, and ethical compliance make it a highly reusable and adapt-able tool for the development of algorithms aimed at improv-ing maternal and Fetal health. (c) 2023 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)