BodyMaps provide a three-dimensional digital atlas of the human body, conceptually similar to Google Maps but focusing on human anatomy. BodyMaps offer three unique features: First, semantic segmentation of anatomical structures, such as the cardiovascular system, skeleton, muscles, and gastrointestinal tract. Second, tumor screening across various structures, including lungs, abdomen, brain, bones, heart, and blood vessels. Third, supporting multiple clinical tasks, such as image registration, disease quantification, tumor stage estimation, report generation, etc. The BodyMaps Project has now contributed a total of 22,682 computed tomography (CT) volumes with detailed, per-voxel annotations of 142 anatomical structures and associated  tumors . This dataset has already attracted 50+ leading research teams worldwide, driving innovation from classical methods (U-Net and its variants) to cutting-edge Foundation Models.

​

In 2009, before the advent of ImageNet, it was challenging to develop artificial intelligence (AI) algorithms that are robust to different domains using a small or even medium size of labeled data. The same situation, we believe, that presents in medical image analysis today, which is in dire need of its own ImageNet moment, where a large amount of data is available, where high-quality annotations are performed, where multiple domains (hospitals) are covered, where the dataset is attached to a widely recognized challenge. The BodyMaps Project presents unprecedented data and annotation scales with over 3.2 million organ/tumor masks and 8.5 million annotated images that are taken from 94 hospitals across 19 countries and manually inspected by expert radiologists.