P4 3D Edge-Aware Refined Atlas Derived from the Allen Developing Mouse Brain Atlas

3D imaging necessitates corresponding 3D anatomical atlases, but generating such atlases is time consuming and challenging. Substantial efforts have generated such atlases including the Allen Developing Mouse Brain Atlas (ADMBA) but are typically limited by the process of serial 2D annotation for 3D reconstruction leading to artifacts. We have developed an automated, edge-aware atlas refinement method to curate existing atlases in a topologically consistent, 3D manner. Applying this approach to the P4 atlas in the ADMBA, we inferred labels for incompletely labeled areas, smoothed the labels in 3D, and realigned the labels to the underlying anatomical edges. As required, we also rotated images in 3D for midline symmetry.

三维成像需要配套的三维解剖图谱（3D anatomical atlases），但此类图谱的构建既耗时又极具挑战。尽管已有大量研究构建了包括艾伦发育小鼠脑图谱（Allen Developing Mouse Brain Atlas，ADMBA）在内的三维解剖图谱，但此类方法普遍受限于用于三维重建的连续二维标注流程，易产生伪影。为此，我们研发了一种自动化、边缘感知的图谱优化方法，能够以拓扑一致的三维方式对现有图谱进行校准整理。我们将该方法应用于ADMBA中的P4阶段图谱，对未完全标注的区域推断其标签，在三维空间中对标签进行平滑处理，并将标签重新对齐至对应的解剖学边缘。同时，我们还按照要求对三维图像进行旋转操作，以实现中线对称。