AI驅動的手術：通過創新和協作實現醫療保健轉型.pdf

《AI驅動的手術：通過創新和協作實現醫療保健轉型.pdf》由會員分享，可在線閱讀，更多相關《AI驅動的手術：通過創新和協作實現醫療保健轉型.pdf（69頁珍藏版）》請在三個皮匠報告上搜索。

1、AI AI-powered Surgery:powered Surgery:Transforming Healthcare Through Transforming Healthcare Through Innovation and CollaborationInnovation and CollaborationAssociate Professor of Surgery,Duke University School of MedicineVice Chair for Innovation,Department of SurgerySurgical Director of Duke AI H

2、ealthDoha,Qatar December 10,2024Ozanan R.Meireles,MD,FACSThe Era of Technological DisruptionGeneral Anesthesia 1840sAntiseptic Surgery 1860s Endoscopic Procedures 1960sCognitive Computing 2010s Surgical AI 2016 Hardware(instruments,Robotics)Software(Data,Algorithms)Human OperatorsVIDEOHuman(professi

3、onal preparedness)Economics(business plan)Governance TelecomEvolution of the OperatingEvolution of the Operating RoomRoomVideo DATA More computing powerMore powerful/efficient techniquesWhat Premier League clubs are learning from F1-The AthleticError Handling and RecoveryError Handling and RecoveryT

4、he average hospital generate137 terabytes per dayF1 teams combined generate243 terabytes per dayOperative planning:Operative planning:Surgical Strategy with Surgical Strategy with Error Handling and Error Handling and Recovery SystemRecovery SystemO.S.Potential Current ApplicationsPotential Current

5、Applications Data-driven insights Predictive analytics Automated scheduling Resource allocation Quality improvement Notification Systems Operative Report Generation Billing Compliance Monitoring Resource Prediction,and AllocationTele-mentoringOperative Case Length PredictionAttention AwarenessBarrie

6、rsData Quality And Privacy ConcernsHigh Implementation CostsComplex IntegrationCultural ResistanceLimited AI InterpretabilityEthical Considerations Relative Advantage Compatibility to Workflow Complexity of Use Triability of models Observability of ResultsDiffusion of Innovation TheoryThree Major Pi

7、llars of Development Establishing a Community of users and developers Building the Architectural Framework to connect and support the community.Establishing a Legal Framework to ensure trustworthy,responsible,ethical,and safe development of artificial intelligence.Annotation Data Structure and Use G

8、overnance Policies,Regulations,and OversightFoundational work Video Data Acquisition Framework Creation of a Community Management through Data LifecycleStructural needs Scientific Research Education Cultural Transformation Knowledge creation and disseminationPreparation,Creation,and ImplementationSu

9、rgical AI StandardsDATA collection ClinicianPatientResearcherIndustrySurgical DATADATA Governance,Policies and OversightHealthcare SystemsGovernmentsSocietiesEthics CommitteesDataConsensus Recommendations on an Annotation Framework for Surgical VideoAnnotation Framework Hierarchical Structure with E

10、xpandable Granularity Temporal Events Phase(generic)Step(procedure-specific)Task(generic)Action(generic)Spatial Events Anatomic region Specific anatomy General anatomy Tissue characteristics Laparoscopic CholecystectomyAccessPort placementPort placementInspectionExecutionExposureRetractionDissection

11、 of the gallbladder HilumOpening of peritoneumDissectionIsolation,ligation,and division of structuresDissectionClippingCuttingRemoval from liverSeparation of the peritoneumRemoval of specimen,hemostasisBaggingHemostasisClosureRemoval of portsPort removalTemporal HierarchyAnnotationAnnotationSpatial

12、HierarchyStructureGovernance UseExplorationUse Cases CommunityScientific efforts Discovery Validation and BenchmarkingClinical TrialsComputer Vison Challenges Multi-institutional collaborationsStandards for PublicationsAcademia and Industry partnershipValidation Studies Promote Diversity Computer Vi

13、sion Competition for Model Benchmarking DATA DonorsAccounts:155Institutions:6429%LMIC1700+USA34%Brazil16%UK7%Australia7%Spain6%Israel6%India5%Italy4%Germany4%France2%Iran1%Peru1%Canada1%Pakistan1%Thailand0%Casablanca0%Vietnam0%Ecuador0%Slovenia0%Greece0%1500 San Pablo St0%Malta0%Egypt0%Japan0%Guatem

14、ala0%Cyprus0%Azerbaycan0%Latvia0%China0%Polska0%10.000 Unique Visitors The SAGES Critical View of Safety Challenge EckhoffIndustry AdvisorsMedTech/Surgical MIS Companies1 Data Anonymization Pipeline5 Sponsors Annotators19 Residents&Fellows14 Countries20%Dropout RateData Donors55 Surgeons54 Instituti

15、ons24 CountriesSUPPORTERSCommunity EffortORGANIZERSSurgeons6 Clinical Leaders14 Clinical AdvisorsComputer Scientists7 Technical Leaders8 Technical AdvisorsSAGESThe SAGES Critical View of Safety Challenge EckhoffClinical Diversity(besides origin)9 Recording DevicesRobotics/LaparoscopicICG+/-IOCTechni

16、cal Diversity19 Resolutions4 File Types3 FormatsOutreach:Data Characteristics3D Vision,HD-Resolution,Robotics+ICG2D Vision,854x480 Pixel,Laparoscopic The SAGES Critical View of Safety Challenge Eckhoff1)CVS ClassificationBinary classification of 3 visually distinct criteriaChallenge Goals:Subchallen

17、ges2)RobustnessConsistency across distribution shifts linked to clinical&technicaldiversity2)Uncertainty QuantificationAccounting for inherent uncertainty in real-world clinical settingThe SAGES Critical View of Safety Challenge Eckhoff36 Teams16 CountriesThe SAGES Critical View of Safety Challenge

18、EckhoffRankingsCVS ClassificationTeam Farm Stanford UniversityTheator IncSDS-HD German Cancer Research CentermmllTUE-VCAPandasFightTumorCaresyntaxSRV-WEISSIRCV-URVTransformersCVS_HUSTHFUT-MedlAUncertainty QuantificationTheator IncPandasChengdu University ChinaSDS-HDGerman Cancer Research CenterSRV-W

19、EISSTransformersCVS_HUSTmmllTUE-VCAFarmCaresyntaxIRCV-URVFightTumorHFUT-MedlASC2-RankingSDS-HDGerman Cancer Research CenterTheator IncTeam Farm Stanford UniversitymmllTUE-VCAPandasCaresyntaxFightTumorIRCV-URVSRV-WEISSTransformersHFUT-MedlACVS_HUSTThe SAGES Critical View of Safety Challenge EckhoffWi

20、nners OverallWinnersTheator IncSDS-HD German Cancer Research CenterTeam Farm Stanford UniversityOverall Prize Money:$15.500+NVIDIA Developer Kit IGX Origin with RTX 6000 ADA GPU (approx.$20k)The SAGES Critical View of Safety Challenge EckhoffCVS Challenge Part 2CVS Challenge Part 2Scientific Meeting

21、sDedicated Fellowships e.g.SAIILMedical School Curriculum PublicationsEducation and TrainingEducation and TrainingScientific MeetingsDedicated Fellowships e.g.SAIILMedical School Curriculum PublicationsEducation and TrainingData governance is a principled approach to managing data during its life cy

22、cle,from acquisition to use to disposal.Surgical AI Governance StakeholdersRegulations,Policies and OversightLegal FrameworkLegal FrameworkTRAIN-SAIILBlueprint WardleyWardley Map Map involves mapping components of a system or process based on their value chain and their stage of evolution.Building t

23、he Surgical Operating System(S.OS)Ethical and Trustworthy Data Generation,Model Development,and ValidationAddressing the Critical Need for Benchmarking and Ethical ConsiderationsData PrivacyAI Model DevelopmentValidationGovernance18 Sep 2024P5862-P-002 v1.0Commercially Confidential4418 Sep 2024P5862

24、-P-002 v1.0Commercially Confidential4518 Sep 2024P5862-P-002 v1.0Commercially Confidential46User:.Password:.SS.OS is a conceptual framework that aims to seamlessly integrate surgical teams,operating rooms,patient data,and devices.StandardizationEfficiencySafetyS.OS Features Efficiency&SchedulingData

25、&Safety ManagementTechnological IntegrationUser Interface&ExperienceSecurity&Access ControlCommunication&CollaborationAnalytics&MonitoringUtilities&Special FeaturesData Generation and ManagementOrganizing Critical InformationMaintenance of structure dataModel Development Data trainingAlgorithm optim

26、izationModel testingContinuous learning and improvementTrustworthiness of AI Systems Assurance:Thorough validation Benchmarking Ensure AI systems are accurate,safe,and ethical.Solution ManagementSecurity and Access ControlEnsuring Data IntegrityMarketplace for Cognitive Augmentation Information Guid

27、ance Safety Operational Efficiency S.OS Surgical Fingerprint Sleeve GastrectomyNormalized Cumulative log ProbabilityCase ACase BSLEEVEnetSLEEVEnetFuture Steps Surgical Video Foundation ModelsThese models serve as a fundamental base,trained on large datasets,and can be adapted to a variety of surgica

28、l tasks such as:Video analysis Complication prediction Real-time guidance AutomationLaparoscopic CholecystectomyAccessPort placementPort placementInspectionExecutionExposureRetractionDissection of the gallbladder HilumOpening of peritoneumDissectionIsolation,ligation,and division of structuresDissec

29、tionClippingCuttingRemoval from liverSeparation of the peritoneumRemoval of specimen,hemostasisBaggingHemostasisClosureRemoval of portsPort removalTemporal HierarchyAnnotationSpatial HierarchyTransfer LearningAI Models BenchmarkingSustainability(ROI)Data monetization Model co-development Application

30、 subscriptions Contracting clinical trials Pre and Post market evaluationSurgical AIStructureGovernance UseExplorationSurgeonsGovernment officialsHospitalsPayersMalpractice Insurance companiesIndustryPatientsSurgical Event Real Time PredictionSUPR-GAN:SUrgical PRediction GAN for Event Anticipation i

31、n Laparoscopic and Robotic Surgery Y.Ban,G.Rosman,J.A.Eckhoff,T.M.Ward,D.Hashimoto,T.Kondo,H.Iwaki,O.Meireles,D.Rus.IEEE Robotics and Automation Letters(RA-L)/ICRA 2022:Ground Truth:PredictionError Handling and RecoveryError Handling and RecoveryAnalysis of intraoperative video with Decision Support

32、.Case 3 Preventing ComplicationConceptual PhaseNormal rangeDeviationWarning!Ozanan Meireles,MDDirector,MGH SAIILDaniela Rus,PhDDirector,MIT CSAILGuy Rosman,PhDAssoc Director,EngineeringDaniel Hashimoto,MD MSFormer FellowThomas Ward,MDFormer FellowYutong Ban,PhDFormer FellowJennifer Eckhoff,MDFormer FellowAlumniFaculty and FellowsLianhao Yin,PhDPostdoctoral FellowDirector of Analytics and InnovationSAIIL TeamThank you!Thank you!GET INVOLVED www.SAIIL.orgOzanan.MeirelesDuke.edu