起步顺序Order of work 先给所有模型建清单、分级;再按级别培训;最后才把运营类模型引入全院。List and tier every model first, then train staff by tier, then bring operations models in.
公开事实Public facts 已核对Checked
- 18M心血管影像模型的训练视频数Videos used to train its cardiac imaging model来自 30 万例 UHN 研究From 300,000 UHN studies
- 20%关键测量误差降低Lower error on key measurements无需再训练即可解读儿科扫描Can read paediatric scans without retraining
- 1.5M用于训练公开心律模型的心电图数量ECGs used to train a public heart-rhythm model供全球研究者使用Available to researchers worldwide
- 18LIGAND-AI 药物发现项目的合作方数量Partners in LIGAND-AI, an A I drug discovery project来自 9 个国家;与 SGC、SickKids 牵头In 9 countries; with SGC and SickKids
模型分级,规则先定Tier every model, fix rules in advance 判断Judgment
- 一级 · 研究模型Tier 1 · Research models:建清单;点名负责人Inventory; a named owner
- 二级 · 运营模型:床位、手术室、急诊流Tier 2 · Operations models: beds, theatres, emergency flow:基线与对照;由人确认Baseline and control; a person confirms
- 三级 · 临床使用模型:影像、心电图、计划Tier 3 · Clinical-use models: imaging, ECG, planning:写清真实目标与代理指标;按患者群体看结果Real target beside the proxy; results by patient group
专家组裁决Expert panel 依据公开观点的推演Inference from public views
Ziad Obermeyer
每个模型:它预测什么、你要什么、按人群看结果。For each model: what it predicts, what you want, results by group.
Thomas Davenport
运营指挥中心改变的是床位与手术室的运转方式。Operations command centre changes how beds and theatres run.
Ethan Mollick
让临床人员在日常工作中测出模型失败的地方。Let clinicians test where models fail, in daily work.
Erik Brynjolfsson
预算含培训与流程再造;对照基线计收益。Budget for training and redesign; count baseline gains.
Andrew Ng
选一个病区流程,三到六个月出数字。One ward flow with numbers in three to six months.
Foteini Agrafioti
AI Hub、各院区与运营的权责写成一页纸。One page of ownership: AI Hub, sites and operations.
Andrew Ng:快速试点一个病区流程。Pilot one ward flow fast.
Ziad Obermeyer:先建模型清单。Inventory models first.
→ 运营走快速通道;临床走完整审查。Fast lane for operations; full review for clinical.
六位专家是谁、怎样思考、有什么铁律 →Who the six experts are, how they think, their rules →
三个目标Three priorities 建议Recommendation
给每个模型建清单、分级List and tier every model
覆盖 10 个站点的一张清单,写明目标、代理指标与负责人。One inventory across 10 sites, with target, proxy and owner.
按级别培训员工Train staff by tier
以 Michener 证书课程为基础,覆盖临床与行政人员。Build on the Michener certificate for clinicians and admin staff.
起步运营指挥中心Start an operations command centre
床位、手术室与急诊流,先做一个流程。Beds, theatres and emergency flow, one flow first.
场景组合Scenarios 建议Recommendation
| 场景Scenario | 做法Approach | 状态Status |
|---|---|---|
| 全院运营指挥中心Hospital operations command centre 提效Efficiency体验Experience | 预测床位、手术室与急诊分流;先从一个流程起步。Forecast beds, theatre slots and emergency triage; start with one flow. | 建议 · 基于 AI HubProposed · builds on AI Hub |
| 跨院模型验证Cross-hospital model validation 降风险Risk体验Experience | 与合作方共同验证模型;数据主权规则与隐私顾问确认。Test models with partners; confirm data-sovereignty rules with privacy counsel. | 建议 · 隐私审查Proposed · privacy review |
| 分层 AI 培训Tiered A I training 提效Efficiency | 按岗位的短课程,让更多员工会用、也敢质疑模型。Short courses by role, so more staff can use and question models. | 建议 · 基于 Michener 课程Proposed · builds on Michener course |
| 向其他医院输出方法Method export to other hospitals 增收Revenue体验Experience | 输出分级与审查方法,而不是患者数据。Share the tiering and review method, not patient data. | 建议Proposed |
12 个月路线12-month roadmap 建议Recommendation
奠基Foundation
- 盘点所有模型Inventory every model
- 风险分级与负责人Risk tiers and owners
- 一页权责One page of ownership
试点Pilot
- 一个病区流程预测One ward flow forecast
- 基线与对照Baseline and control
- 临床人员测边界Clinicians test the edges
上线Go live
- 分层培训启动Tiered training starts
- 流程上线Flow goes live
- 上线即监控Monitoring from day one
扩展Extend
- 加入第二个流程Add a second flow
- 分享方法Share the method
- 重排下年优先级Re-rank next year
治理与合规Governance 建议Recommendation
- 患者数据与共享:规则与隐私顾问确认Patient data and sharing: confirm rules with privacy counsel
- 临床模型:使用前经临床治理审查Clinical models: clinical governance review before use
- 影响诊疗的输出由临床人员确认A clinician confirms any care-affecting output
三层指标Three-layer metrics 建议Recommendation
| 采用Adoption | 按级别与岗位统计的受训员工数Staff trained, by tier and role · 有负责人与分级的模型数Models with an owner and tier |
|---|---|
| 过程Process | 预测准确度对照基线Forecast accuracy against baseline · 临床人员确认率Clinician confirmation rate · 完成一次模型评审的时长Time to assemble a model review |
| 结果Outcome | 床位等待时间对照基线Bed wait times against baseline · 按患者群体比较结果Results compared across patient groups |
目标值由企业自定。Targets are the company's to set.
风险Risks 判断Judgment
模型没错,目标选错Right model, wrong target
模型可能预测的是代理指标(如费用)而非需求;两者都写明。A model can predict a proxy, such as cost, instead of need; name both.
工具强,会用的人少Strong tools, few skilled users
若临床人员不会检验,模型会被闲置。Models sit unused if clinicians are not trained to test them.
人数口径不一Mixed headcount figures
不同来源的人数口径不同;规划培训时统一定义。Staff counts differ by source; use one definition when sizing training.
前 90 天First 90 days 建议Recommendation
| 第 1–2 周Weeks 1–2 | 为每个站点点名清单负责人Name inventory owner at each site |
|---|---|
| 第 3–8 周Weeks 3–8 | 列出每个模型、目标与代理指标List every model, target and proxy |
| 第 9–12 周Weeks 9–12 | 起草分级;选一个病区流程Draft tiers; choose one ward flow |
| 第 13 周Week 13 | 首次复盘First review |
我们可能错在:We could be wrong if: 临床人员认为分级拖慢了有用的研究,或病区流程相对基线没有收益。clinicians find tiering slows useful research, or the ward flow shows no gain over baseline.
出处Sources
本页是基于公开资料的分析,与所述企业没有合作关系,不代表其官方立场或内部计划;企业数字多来自其自述或媒体转述;「专家组」是依据六位专家公开观点的推演,不是他们的原话,也不代表他们或其雇主的立场。资料整理日期:2026-11-14。This page is an analysis based on public information. It has no affiliation with the company and does not represent its official position or internal plans. Company figures come mostly from its own statements or media reports. The expert panel is an inference from six experts' public views, not their words, and does not represent them or their employers. Compiled 2026-11-14.
想给你的公司也做一份?Want one for your company?
同样的方法:公开事实、专属判断、12 个月路线和前 90 天。公司大小都可以聊,中文沟通。Same method: public facts, a company-specific judgment, a 12-month roadmap and the first 90 days. Any size of business; we can talk in Chinese or English.