Muro Satoru, Ibara Takuya, Nimura Akimoto, Akita Keiichi
Department of Clinical Anatomy, Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo, Japan.
Department of Functional Joint Anatomy, Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo, Japan.
Microscopy (Oxf). 2025 Mar 3. doi: 10.1093/jmicro/dfaf015.
Traditional three-dimensional reconstruction is labor-intensive owing to manual segmentation; this can be addressed by developing artificial intelligence-driven automated segmentation. However, it is limited by a lack of user-friendly tools for morphologists. We present a workflow for three-dimensional reconstruction using our artificial intelligence-powered segmentation tool. Specifically, we developed an interactive toolset, "Seg & Ref," to overcome the abovementioned challenges by enabling artificial intelligence-powered segmentation and easy mask editing without requiring a command-line setup. We demonstrated a three-dimensional reconstruction workflow using serial sections of a Carnegie Stage 15 human embryo. Automated segmentation (Step 1) was performed using the graphical user interface, "SAM2 GUI for Img Seq," which utilizes the Segment Anything Model 2 and supports interactive segmentation through a web-based interface. Users specify target structures via box prompts, and the results are propagated across all images for batch segmentation. The segmentation masks were reviewed and corrected (Step 2) using "Segment Editor PP," a PowerPoint-based tool enabling interactive mask refinement. Finally, the corrected masks were imported into the 3D Slicer application for reconstruction (Step 3). Our three-dimensional reconstruction visualized key structures, including the spinal cord, veins, aorta, mesonephros, gut, heart, trachea, liver, and peritoneal cavity. The reconstructed models accurately represented their spatial relationships and morphologies. This provides a labor-saving approach for three-dimensional reconstruction workflows owing to their optimization for serial sections, versatility, and accessibility without programming expertise. Therefore, morphological research can be enhanced by precise segmentation using intuitive and user-friendly interfaces of "Seg & Ref."
传统的三维重建由于手动分割而劳动强度大；这可以通过开发人工智能驱动的自动分割来解决。然而，它受到形态学家缺乏用户友好工具的限制。我们展示了一种使用我们的人工智能驱动的分割工具进行三维重建的工作流程。具体来说，我们开发了一个交互式工具集“Seg & Ref”，通过实现人工智能驱动的分割和无需命令行设置的轻松掩码编辑来克服上述挑战。我们使用卡内基15期人类胚胎的连续切片展示了一个三维重建工作流程。自动分割（步骤1）使用图形用户界面“SAM2 GUI for Img Seq”进行，该界面利用了Segment Anything Model 2并通过基于网络的界面支持交互式分割。用户通过框选提示指定目标结构，结果会在所有图像中传播以进行批量分割。使用“Segment Editor PP”（一个基于PowerPoint的工具，可实现交互式掩码细化）对分割掩码进行审查和校正（步骤2）。最后，将校正后的掩码导入到3D Slicer应用程序中进行重建（步骤3）。我们的三维重建可视化了关键结构，包括脊髓、静脉、主动脉、中肾、肠道、心脏、气管、肝脏和腹膜腔。重建模型准确地呈现了它们的空间关系和形态。由于其针对连续切片进行了优化、具有通用性且无需编程专业知识即可访问，这为三维重建工作流程提供了一种省力的方法。因此，通过使用直观且用户友好的“Seg & Ref”界面进行精确分割，可以加强形态学研究。
