01
Position定位
Is the clip on target?夹子对准目标了吗？
Today现在Estimated from two planes by eye.凭肉眼从两个切面估计。
AutoClipClip trajectory onto the target, in both planes.夹子在两个切面上相对目标的轨迹。
AutoClip· First indication: mitral TEER· 首个适应症：二尖瓣 TEER
Real-time AI reads the heart and guides repair. 实时 AI 读懂心脏，引导瓣膜修复。
During a catheter valve repair the operator steers by live ultrasound, and every decision is made by eye. AutoClip reads that image and gives the operator a measurement where there was a guess. 经导管瓣膜修复中，术者靠实时超声操作，每一个决策都凭肉眼。AutoClip 读懂这幅图像，在原来靠猜的地方给术者一个测量值。
TCT 2025
Featured Innovation, San Francisco, while still a prototype旧金山 Featured Innovation，原型阶段即入选
~2,000
TEER procedures from four centres, under data-use authorizations例 TEER，来自四个中心，经数据使用授权
≈1.5 mm
clip-position error against expert labels, 34 held-out patients夹子位置与专家标注的误差，34 位留出验证患者
12 / 12
centres shown the demo asked to join external validation看过演示的中心全部要求参加外部验证
01The problem问题
Every TEER decision is still made by eye.TEER 的每一个决策，仍然靠肉眼。
Where to place the clip, when to grasp, whether to release. The operator reads a grey, moving image, and there is not a single number on the screen. The result decides the patient's outcome, and the result depends on who operates. 夹子放在哪、何时抓、能不能释放。术者读的是一幅灰色的动态图像，屏幕上没有一个数字。结果决定患者的结局，而结果取决于谁在做。
2–4mm
disagreement between four experienced specialists on the same frame and the same landmark: the papillary muscle that decides clip position.四位资深专家在同一帧、同一解剖标志上的分歧。这个标志是决定夹子位置的乳头肌。
Turicor inter-observer studyTuricor 观察者间一致性研究38% vs 72%
death or heart-failure hospitalisation by two years, when residual leak after the procedure is minimal (≤1+) versus severe (≥3+).术后残余反流轻微（≤1+）与严重（≥3+）的患者，两年内死亡或心衰住院的比例。
COAPT · Kar et al., Circulation 202164% → 75%
optimal procedural success from an operator's first 25 cases to 50+. Most centres never reach fifty cases a year.术者前 25 例到 50 例以上的最佳手术成功率。多数中心一年做不到 50 例。
STS/ACC TVT Registry · Chhatriwalla et al., JACC 201902AutoClip
A number where there was a guess.原来靠猜的地方，现在有一个数字。
AutoClip segments 15 structures on every frame of the live echo (leaflet scallops, grippers, insertion points, papillary muscles, the clip) and turns them into a measurement for each of the three decisions. Decision support, not decision replacement: the operator stays in control. AutoClip 在实时超声的每一帧上分割 15 个结构（瓣叶分区、抓钳、插入点、乳头肌、夹子），把它们变成三个决策各自的测量值。辅助决策，不替代决策：控制权始终在术者手里。
02
Grasp夹持
Are both leaflets inside?两侧瓣叶都夹进去了吗？
Today现在Judged on a moving image.在动态图像上判断。
AutoClipInsertion depth of each leaflet inside the clip, in millimetres.每侧瓣叶在夹子内的插入深度，以毫米计。
03
Release释放
Is the result good enough to let go?结果够好，可以释放了吗？
Today现在A judgement call, case by case.逐例凭经验判断。
AutoClipFive release criteria, each checked before letting go.五项释放标准，逐项核对后再释放。
03In the cath lab在导管室里
Nothing in the room has to change.导管室里什么都不用改。
In输入
Echo machine超声机
The video output it already has. Any vendor.它本来就有的视频输出。任何厂商。
AutoClip
A GPU box in the lab导管室里的一台 GPU 盒子
Reads every frame. No hospital network, no cloud.逐帧读图。不接医院网络，不上云。
Out输出
A second screen第二块屏幕
Measurements next to the live image the operator already watches.测量值显示在术者本来就在看的实时图像旁边。
Any clip任何夹子Device-agnostic. Built first for the clip with most of the market.不绑定器械。首先适配市场份额最大的夹子。
Any echo vendor任何超声厂商Works from the standard video signal, not a vendor interface.读取标准视频信号，不依赖厂商接口。
Today: offline replay现在：离线回放Running on recorded procedures now. Real-time on the live feed is the next build.目前在录像病例上运行。接实时视频流是下一步。
04Evidence证据
The model is only as good as the clinical definition.模型的上限，是临床定义的质量。
Our advantage is not generic computer vision. It is specialist knowledge turned into labels and protocols, trained on in-domain data, and tied to what happened to the patient.我们的优势不是通用计算机视觉，而是把专科知识变成标注和规程，用术中数据训练，并与患者的真实结局挂钩。
15
structures defined on every frame, by operators个结构，由术者定义，逐帧标注
235
patients annotated, ~10,000 frames, under a written protocol例患者已标注，约 10,000 帧，按书面规程
4
annotators cross-checked: the expert ceiling was measured before the model位标注者交叉核对：先测专家上限，再评模型
203
procedures linked to registry follow-up, to test which readouts predict outcome例手术关联登记库随访，检验哪些读数能预测结局
Done已完成
- 15-structure segmentation and clinical readouts on recorded procedures15 结构分割与临床读数，在录像病例上运行
- Held-out benchmark: 34 patients, two centres, no overlap with training留出验证：34 位患者、两个中心、与训练零重叠
- Outcome-linked cohort of 203 intra-procedural studies203 例术中检查的结局关联队列
Next · by March 2027下一步 · 2027 年 3 月前
- Real-time inference on the echo video output在超声视频输出上实时推理
- Shadow mode in two cath labs under a research protocol两个导管室按研究方案做影子模式运行
- FDA Pre-Submission on pathway and validation plan就路径与验证方案向 FDA 提交 Pre-Submission
Not yet尚未
- Use during patient care用于患者诊疗
- External validation外部验证
- Regulatory clearance of any kind任何监管许可
05Why now为什么是现在
The third generation of edge-to-edge repair.缘对缘修复的第三代。
1991 · Surgical外科
The Alfieri stitchAlfieri 缝合
One suture between the free edges of the two leaflets, Milan. A simple, reproducible repair.米兰。一针缝合两侧瓣叶的游离缘，简单、可重复的修复。
2003 · Transcatheter经导管
The clip, first-in-human夹子首次用于人体
The same repair delivered through a catheter and judged on echo. 300,000+ patients; guideline therapy.同样的修复经导管完成，在超声上判断。30 万+ 患者，写入指南。
2026 · AI-guidedAI 引导
AutoClip
The machine learns to see the repair on the echo image. Every decision made by eye gets a number.机器学会在超声图像上看懂这台修复。每个靠肉眼的决策都有了数字。
06Team团队
An operator and a computer-vision scientist.一位术者，一位计算机视觉科学家。
Founder & CEO创始人 · CEO
Mi Chen, MD PhD 陈密
- Cardiac surgeon and TEER / TAVI operator; 1,700+ structural heart procedures as operator and team member心脏外科医生，TEER / TAVI 术者；主刀与参与结构性心脏手术 1,700+ 例
- Trained at Fuwai Hospital, Beijing; structural heart fellowship at University Hospital Zürich and Klinik Hirslanden北京阜外医院训练；苏黎世大学医院与 Hirslanden 医院结构性心脏病 fellowship
- Faculty at TCT, EuroPCR and PCR London ValvesTCT、EuroPCR、PCR London Valves 讲者
- Writes the model code herself模型代码由她本人编写
Co-founder & CTO联合创始人 · CTO
Denis Samuylov, PhD
- PhD in computer vision, ETH Zürich; a decade in the field苏黎世联邦理工学院计算机视觉博士，深耕十年
- Co-founder of GenAI Zürich, the applied-AI event series应用 AI 系列活动 GenAI Zürich 联合创始人
- Owns the architecture and the real-time inference pipeline负责系统架构与实时推理管线
“AutoClip is the tool I wanted on the screen every time I clipped a valve.”「AutoClip 是我每次夹瓣膜时，都希望屏幕上有的那个工具。」 Mi Chen
07Where this goes方向
The intelligence layer of cardiac medicine.心脏医学的智能层。
We start with the mitral clip. The goal is AI that understands the beating heart in real time, for every catheter procedure. The operating room comes first because it is where perception and action already meet, and every case closes the loop.我们从二尖瓣夹子开始。目标是能实时理解跳动心脏的 AI，服务每一台导管手术。先从手术室做起，因为那里感知与动作已经相遇，每一台手术都让闭环转一圈。
- Now现在Mitral TEER二尖瓣 TEER · one workflow, validated at outside centres一个工作流，在外部中心验证
- Next下一步Tricuspid TEER, TAVI三尖瓣 TEER、TAVI · one shared heart model across procedures跨术式共用一个心脏模型
- Later之后Beyond the valve瓣膜之外 · only where the evidence carries over只在证据能迁移的地方
08Contact联系
Build it with us.和我们一起做。
Operators & centres术者与中心
Validate the live version验证实时版本
We are choosing two cath labs for research-only shadow-mode runs, in Europe and the US.我们正在欧洲和美国选两个导管室，做仅用于研究的影子模式运行。
hello@turicor.com → Device makers器械厂商Evidence for your clip为你的夹子积累证据
Imaging AI and outcome evidence for clips that ship without it.为没有影像 AI 的夹子提供 AI 与结局证据。
hello@turicor.com → Engineers工程师Real-time imaging, SF实时影像，旧金山
ML and real-time imaging engineers who want their code in the operating room.想让自己的代码进手术室的 ML 与实时影像工程师。
hello@turicor.com → Investors投资人Talk to the founders和创始人聊
We are in the San Francisco Bay Area and happy to show the demo.我们在旧金山湾区，欢迎当面看演示。
hello@turicor.com →