智慧医疗：台湾的机会和挑战.pdf

智慧醫療:台灣的機會和挑戰李建璋台大智慧醫療中心副主任台大急診部主治醫師史丹佛大學 Biodesign 教學研究員哈佛大學流行病學博士Automatic interpretation of radiological imagesMonitoring disease risksDevelop new drugsPredict COVID trendIt seemed that Watson could understand the meaning of language,rather than just patterns of wordsWatsons Medical AI Moon Shot ProjectWatsons electronic brainDr.WatsonDr.Watson Goal:Reduce diagnosis errors Optimize treatments Alleviate doctor shortages How Collection of symptoms and came up with a list of diagnoses and treatment Each annotated with Watsons confidence level and links to literature Confidence Memory banks held knowledge of every rare disease and newest trials Processors not susceptible to cognitive bias“This product is a piece of sh.”“Multiple examples of unsafe and incorrect treatment recommendations,”Its all about“DATA”Its all about“DATA”That narrative text accounts for about 80 percent of a typical patients recordand its a stew of jargon,shorthand,and subjective statements Even for MD Anderson,they only have hundreds of beds,could not cover the vast variety of human diseases Trained the software on a small number of hypothetical cancer patients,rather than real patient data leads to disasterDirty Medical DataDirty Medical Data Heterogeneity Different drug preference for different hospital/physicians Different treatment recommendation over different periods Different insight for different levels of doctors Different selection of testing orders Problems specific for Taiwan Use a lot of atypical short-hand Physicians note,not English,not Chinese,but“Taiwanese Chinglish”Under-recording for doctors note,overrepresenting of nurse charting notes Of 29 FDA approved SamD,Image applications account for 70%台灣的機會和挑戰Lets build a Medical ImageNet of TaiwanAI model is ethnic/region sensitiveAI model based on Western countries may not be able to apply to Asian countries台灣三座健保資料庫電病歷資料庫體物資料庫挑戰-Privacy Issue26資料ChallengeOpportunityMedical image needs lots of efforts for annotationCross-disciplinary Training30才ChallengeOpportunityAnnotationISTCross-disciplinaryprofessionals 醫學影像辨識、基因體分析需要龐大的電腦儲存與算力醫療影像:提升2倍效能自然語言:提升5倍效能GPU 分成七個執行體，提升7倍效能支援基因高速運算，提升2.25倍效能人類全基因定序僅需20分鐘臺大醫院建置二套 NVIDIA DGX-A100 目前運算效能最快的AI超級電腦33算ChallengeOpportunityStrong IT industryUnderuse of cloud computingAI paper hits record highNewly funded AI startup record lowAI 模型需要第三期臨床試驗資料資料華人最優質的醫療大數據人才人才算力算力轉譯轉譯60個臺灣大學人工智慧團隊/啟用2台AI 超級電腦/建置運算專區臨床試驗場域/推動醫療醫材軟體試用台大醫院成立智慧醫療中心台大醫院成立智慧醫療中心3940病患隱私資料不外洩落地辦法確效機制與定期紀錄資訊安全不影響Portal系統效能台大醫院醫療器材軟體試用申請流程43轉譯ChallengeOpportunityEducationRegulationHigh Quality Clinical Trial CenterPHASE III TRIAL44DATADATA SCIENCEDATA ECONOMYPRIVACYREGULATIONThank You