围绕 LIBR 软组织变形矫正算法，我们建立了一套涵盖感知、计算、协作的完整 Surgical Spatial AI 平台。Around the LIBR deformation correction algorithm, we've built a complete Surgical Spatial AI platform spanning perception, cognition, and collaboration.
软组织手术导航最难的一步，就是在术中实时校正"术前影像和真实组织对不上"的偏差。这个问题十年没人真正解决 —— 我们用 LIBR 算法做出了行业里第一个能跨越临床可用线（5mm 以内）的方案。The hardest problem in soft-tissue navigation is real-time intraoperative correction — preop imaging never matches the deformed reality. For a decade, no one has truly solved it. LIBR is the first published method to cross the 5mm clinical usability threshold.
自研外部反射参考工具技术，独立于头显惯导系统的固定世界坐标系 —— 即使医生大幅活动，全息内容保持毫米级稳定。Proprietary external reflective reference tools establish a world coordinate system independent of headset IMU drift — holograms stay millimeter-stable even during heavy surgeon motion.
基于头戴 AHAT 深度相机和自研算法的无标记、亚毫米级跟踪 —— 无需额外硬件即可获得高精度数据。Markerless sub-millimeter tracking via headset AHAT depth + proprietary algorithms — no extra hardware needed.
SAM 2 自动分割 + 三路重建融合（原生深度流 / Depth Anything V3 / 非刚性融合）—— 业内首个 Magic Leap 2 上完整跑通的术中肝表面流水线。SAM 2 auto-segmentation + tri-path reconstruction (native depth / Depth Anything V3 / non-rigid fusion) — the first complete intraoperative liver surface pipeline on Magic Leap 2.
LIBR / LIBR+ 算法：肝脏 10.4 → 6.1mm（改善 40%+），乳腺 10.4 → 4.2mm，仿体测试 1.7mm —— 行业唯一跨越 5mm 临床可用线的方案。LIBR / LIBR+: liver 10.4 → 6.1mm (40%+ improvement), breast 10.4 → 4.2mm, phantom tests 1.7mm — the only published method below the 5mm clinical line.
KUKA LBR Med / KINOVA Gen3 Med + NDI Polaris Lyra + ZED 2i 多传感器末端。机械臂从"被动设备"变成"主动观察者"，自动重定位、自主避让医生手部。KUKA LBR Med / KINOVA Gen3 Med + NDI Polaris Lyra + ZED 2i sensor head. Transforms the cobot from a passive device into an active observer that auto-repositions and avoids surgeon hands.
乳腺切除后追踪笔扫描标本切面 → 实时切缘距离评估（无需病理科）。肝脏微波消融实时探针轨迹 + 距靶点测量 + 消融区预测可视化。Breast: track-pen surface scan → real-time margin assessment without pathology lab. Liver MWA: live probe trajectory + target distance + ablation zone prediction.
超声从"远端屏幕 2D 切面"升级为"叠加于解剖正确位置的原位三维影像" —— HoloLens 2 上 ~2mm 跟踪精度 + < 16ms 延迟。Ultrasound goes from a 2D slice on a remote screen to an in-situ 3D image at the correct anatomical position — ~2mm tracking accuracy and <16ms latency on HoloLens 2.
在已发表的肝脏与乳腺导航工作之上，我们正把整套流水线推向"自成一体、去硬件、全自动"—— 更少的术中硬件，更快的部署，更低的临床采纳门槛。Building on our published liver and breast navigation work, we are pushing the entire pipeline toward "self-contained, hardware-free, fully automatic" — less intraoperative hardware, faster deployment, and a lower barrier to clinical adoption.
把此前依赖立体相机、外部光学追踪与独立工作站的乳腺影像引导流水线，整合进一台可穿戴设备。同一 RGB-D 数据流采集皮肤标志物与部分乳腺表面点云，驱动非刚性配准，将术前个体化模型形变对齐后，把皮下肿瘤原位叠加在医生视野中 —— 用于保乳手术的实时切缘导航。Consolidates a breast image-guidance pipeline that previously required a stereo camera, an external optical tracker, and a separate workstation into a single wearable device. One shared RGB-D stream captures labeled skin fiducials and a partial breast-surface point cloud to drive a non-rigid registration, deforming a patient-specific preoperative model to overlay the subsurface tumor in situ — for real-time margin navigation in breast-conserving surgery.
面向开腹肝脏手术的全自动、无标记、学习驱动导航系统（Magic Leap 2）。以头显 SAM 2 自动分割肝表面，用学习式 RGB-D 特征检测替代传统触笔逐点数字化，配合部分暴露训练策略与语音免手交互，将术前 CT 中的肝表面、门静脉/肝静脉与肿瘤实时配准并渲染为头稳定的 AR 叠加。A fully automatic, markerless, learning-based navigation system for open hepatic surgery (Magic Leap 2). Headset-side SAM 2 auto-segments the liver surface, and learned RGB-D feature detection replaces manual stylus digitization — combined with a partial-exposure training strategy and voice-driven, hands-free interaction — to register the CT liver surface, portal/hepatic vessels, and tumor in real time and render them as a head-stabilized AR overlay.