MDDI 演講稿 · 2021-12-03
楊穎儀常任秘書在第四屆新加坡醫療 AI 資料馬拉松暨展覽會上的演講
要點
- • AI 是「時機已到」的平臺技術——已被部署到影像識別、邏輯推理、臨床診斷等領域；新加坡醫療已有區域性應用——含 NCID 用 TTSH/A*STAR 的 AI 讀胸片預測重症肺炎、NUS IDentif.AI 加速 COVID 藥物組合篩選、NUHS 用機器人量體溫。
- • 標杆 SELENA+：SNEC + SERI + NUS Computing 聯合開發的糖尿病視網膜病變 AI——準確率 90%+，已在大多數綜合診所部署、2021 年底覆蓋所有綜合診所、年篩查 12 萬人；授權給本地初創 EyRis（業務覆蓋 20+ 國家與地區）。
- • AI 落地需要「全村合力」——臨床、護理、行政、技術、IT 協同；之前已經手工跑通的遠端視網膜診斷——為 SELENA+ 鋪平了道路。
- • 治理工具：2020 年 1 月《Model AI Governance Framework》第二版；今年 10 月衛生部發布《醫療 AI 指南》（AIHGle），與 HSA 的 AI 醫療裝置監管互補。
- • 本地 AI 初創案例：UCARE.AI 的 AI 賬單預測器——做到了「使用 AI 透明、加密敏感資料、最小化使用」。
完整譯文（繁體中文）
MDDI 英文原文譯文 · 翻譯日期: 2026-05-03
本文已從早期版本的網站遷移過來——格式可能有不一致之處。
早安——尊敬的與會者與朋友們：
感謝邀請我今天發言——這是一場把頂級臨床醫生、資料科學家、創新者聚到一起、應對醫療緊迫問題的活動。我祝賀新加坡國立大學醫學組織（NUHS）、新加坡國立大學（NUS）與麻省理工學院（MIT）「Critical Data」組織這一重要活動。
我有幸在公共服務生涯的多個崗位上與你們的社群合作——在衛生部——與許多人一起開發新加坡國家電子臨床醫療記錄與醫療 IT；推動學術醫學；推進生物醫學研究與創新；現在則共同主持政府對「智慧國與數字經濟」的資助與發展工作。
AI 在醫療中具有大規模部署的巨大潛能
基於這一視角——今天我想和大家分享幾點反思。第一——AI 是一項「時機已到」的平臺技術。人工智慧與機器學習——將被廣泛部署。用於影像識別的 AI——已部署到許多應用領域——包括人臉識別、監控、客戶與情感分析。用於邏輯推理的 AI 也在發展——最有名的展示——是一臺計算機在「AlphaGo」比賽中擊敗了世界頂級棋手。它因此可以被應用到臨床診斷。在新加坡——AI 已在醫療的若干領域被區域性應用——包括目前抗擊 COVID-19 的工作中——
a. NCID（國家傳染病中心）使用了由 TTSH（陳篤生醫院）與 A*STAR 開發的 AI——讀胸片、預測患者發展為重症肺炎的可能性——讓醫生能更及時地分診與干預。
b. AI 也被用來快速評估「對抗 COVID-19 病毒的最有效藥物組合與劑量」。識別一種可用於治療 COVID-19 的藥物——若手工去做——是一項艱苦的任務——既要時間、也要精度——去篩過海量可能的藥物組合。NUS 的「IDentif.AI¹」顯著加速了這一過程——幾天內就給出答案。
c. 智慧機器人也被部署到多種用途。NUHS 用 AI 加持的機器人——幫助測溫與掃描——以確保入口安全；社群照護設施——則用於密切監測患者——以便及時進行臨床與社會心理干預。這對減輕在社群照護設施中照顧大量 COVID-19 患者的醫療專業人士的壓力——非常寶貴。
因此——我很高興看到這麼多人參與本次 Datathon——尤其是當我們的醫療社群已經被 COVID-19 的需求拉得很緊時。這顯示了你們對「更好的患者照護」事業的投入。我相信——這場 Datathon 會激發跨醫療領域 AI 應用的新想法——惠及你們服務的患者。
應用需要多方利益相關方的「系統級支援」
這就引到我的第二點——從我剛談到的「個體創新火花」——拉昇到「系統級視角」。我的第二個反思是——把 AI 創新應用到改善醫療——需要一支多學科團隊來落地——也需要多利益相關方的努力。「這要全村合力」（it takes a village）。讓我用 SELENA+ 作為說明。SELENA+ 是醫療行業的國家級 AI 專案——由「新加坡全國眼科中心」（SNEC）、「新加坡眼科研究院」（SERI）與 NUS 計算學院共同開發——是一個分析眼底掃描、檢測糖尿病視網膜病變（DR）的 AI 演算法。SELENA+ 的準確率達 90% 以上——與專業評級師的表現相當——但用時短得多。
SELENA+ 不是「憑空出現」——它是「遠端診斷影像創新」的最新一章。新加坡的「遠端放射影像」（teleradiology）努力——至少 15 年前就開始了——包括我們的 X 光片在印度被讀、美國的 X 光片由我們讀——以利用時區差。我們逐漸學會了「確信於臨床標準與患者安全」——這是醫療創新成功的關鍵因素。我記得——當時 SNEC 引入了「在綜合診所對患者進行糖尿病視網膜病變遠端診斷」。我們國家級專科中心的眼科專家——能觸達基層醫療系統中更廣泛的患者——這是一大跨越——能比以往更早檢測與治療患者。「早期檢測」對應對糖尿病非常寶貴。讓這件事發生——需要專科中心與基層醫療網路在很多事情上的協作——包括工作流、患者管理、資料傳輸。它要求臨床醫生、護士、行政人員、技術專家與 IT 專業人士走到一起。要解決障礙、構建協作——需要辛勤工作。因為我們已經把這件事「手工跑通」過——所以現在引入 SELENA+ 時變得順滑。「全村合力」加上「持久的決心與承諾」——才能讓創新落地。
SELENA+ 已在大多數綜合診所部署——並將於 2021 年底覆蓋所有綜合診所。預計每年篩查 12 萬名患者——覆蓋全國綜合診所所有的糖尿病眼科篩查。SELENA+ 也已授權給本地初創 EyRis——其業務覆蓋歐盟、中國、東南亞等 20 多個國家與地區。
旅程繼續。我們正在舉辦「新加坡醫療 AI 大挑戰」（AI in Healthcare Grand Challenge）——目標是應對影響新加坡人的主要慢性病——糖尿病、高血壓、高血脂。我期待——研究者、臨床醫生、電腦科學家、資料分析師組成的多學科團隊——為此開發出更多創新方案。
「倫理 AI 與信任」的重要性——以及新加坡的思想領導力
我留給各位的第三點——是「我們對 AI 的採用——必須以『安全、倫理使用』為前提」。我們必須管理一種微妙的平衡——既鼓勵創新——又維持公眾對所部署技術的高水平信任。在醫療領域——這一點尤其重要——因為「誤用」可能給患者帶來巨大傷害——並摧毀信任。
回應「AI 使用傷害風險」的擔憂——需要我們深入思考一些跨政策、法律、技術、倫理的問題——比如——
a. AI 容易受錯誤與偏見影響。我們如何評估「什麼準確度算可接受」——以及它如何在不同用例間變化？
b. 我們如何確保 AI 方案足夠穩健——以防範 AI 演算法的不負責任部署、訓練資料的不當管理？以及如何防範網路安全風險（包括資料操縱）？
c. 一旦 AI 系統出錯——我們如何確定問責？在真實生活的併發情境裡——誰來負責？
這些挑戰在哪裡都適用——但在醫療中尤具迴響——因為它影響患者的健康與生命。所以在你們的討論與創新開發中——請把治理、倫理與韌性的考量——融入到 AI 創新的設計中。
新加坡已採取具體措施——幫助建立原則與實操指南——確保 AI 以負責任、倫理的方式被部署。這對公眾與專業人士對 AI 的信任至關重要。2020 年 1 月——新加坡釋出了《Model AI Governance Framework》第二版——把倫理原則——翻譯為機構能採用的實務措施。今年 10 月——衛生部（MOH）向醫療提供者釋出了《AI 在醫療的指南》（AI in Healthcare Guidelines, AIHGle）——以建議支援患者安全、提升對 AI 的信任——並把最佳實踐分享給 AI 系統開發者與落地團隊。對 AI 醫療裝置製造商——這與新加坡衛生科學局（HSA）對 AI 醫療裝置的監管互補。
醫療行業中的多家公司——已經實施了《Model AI Governance Framework》的條款。舉一個例子——本地醫療初創 UCARE.AI——開發了 AI 賬單預測器——為患者推算精準的住院賬單估值。它在 AI 使用上保持透明——對賬單預測有疑慮的客戶被鼓勵提出；UCARE.AI 也對敏感資料加密、最小化使用——以維護患者保密。據我瞭解——該 AI 方案的部署進展良好——醫院能給出更準確的賬單估值——患者也對賬單大致規模有了更安心的預期。
新加坡也積極參與「倫理 AI」的國際討論。比如——IMDA 與 PDPC 正在開發一個「AI 治理測試框架」——與歐盟、聯合國教科文組織、經合組織（OECD）等國際公認原則一致。AI 的最佳實踐——也出現在我們的《數字經濟協定》中。這些努力支撐新加坡「成為 AI 方案開發、試點、部署與規模化的全球樞紐」的願景。
結尾之前——祝大家有一場富有成效的 Datathon。這幾天專題與工作坊中的對話——把當下醫療議題的一些最新解決方案帶到臺前。我希望——它們能激勵你們——開發出惠及我們醫療系統與你們服務的患者的創新。
再次祝賀 NUHS、NUS 與 MIT Critical Data——讓這一活動成真。祝你們與所有參賽者——一切順利。謝謝。
-------------------------------------------------------------------------------------------------------------------------- 1 Identifying Infection Disease Combination Therapy with AI（用 AI 識別感染疾病的聯合治療方案）
英文原文
MDDI 官網原始記錄 · 抓取日期: 2026-05-02
This article has been migrated from an earlier version of the site and may display formatting inconsistencies.
Good morning, distinguished participants and friends.
Thank you for inviting me to speak today in an event bringing together top clinicians, data scientists and innovators to address pressing problems in healthcare. I congratulate National University Health System (NUHS), National University of Singapore (NUS) and the Massachusetts Institute of Technology (MIT) Critical Data for organising this important event.
I have had the pleasure of working with your community in my various jobs in my public service career – at the Ministry of Health, working with many of you to develop our national electronic clinical medical records and healthcare IT, on developing academic medicine and advancing our work in biomedical sciences research and innovation, and now co-chairing the Govt’s funding and development effort on Smart Nation and the Digital Economy.
AI has huge potential for widespread deployment in healthcare
With this perspective, I’d like to share a few reflections with you today. First, AI is a platform technology whose time has come. Artificial intelligence and machine learning will be widely deployed. AI for image recognition is being deployed in many application areas, including facial recognition, surveillance, customer and sentiment analysis. AI for logic deduction is also being developed, most famously demonstrated by a computer winning an AlphaGo contest against one of the best players in the world. It can therefore be applied to clinical diagnostics. In Singapore, AI is being applied in pockets in healthcare, including in the current fight against Covid-19:
a. NCID used AI developed by TTSH & A*STAR to read chest-x-rays and to calculate the likelihood that a patient may develop severe pneumonia. This has enabled more timely intervention from clinicians to better triage patients.
b. AI has been used to quickly evaluate the most effective combinations and dosage of medicines to combat the Covid-19 virus. Identifying a drug that can be used to treat Covid-19 is a grueling task, if done manually. It would require both time and precision to sieve through the massive range of possible drug combinations for effective solutions. NUS’ IDentif.AI1 has accelerated this process significantly, yielding answers within just days.
c. Smart bots are also being deployed for various uses. NUHS is using AI-enabled robots to help temperature taking and scanning for safe entry. Community care facilities for close monitoring of patients to enable timely clinical and psychosocial interventions. This has been valuable to reduce the strain on our healthcare professionals caring for many Covid-19 patients in our community care facilities.
I am therefore happy to see so many people participating in this Datathon, and all the more so when our healthcare community is already so heavily stretched by the demands on you from Covid-19. It shows your dedication to the cause of better patient care. I am confident that the Datathon will spark new ideas for the potential uses for AI across the spectrum of healthcare, which will benefit the patients you serve.
Application requires system-level support from multiple stakeholders
This brings me to my second point, which zooms up from the individual innovative spark I just spoke of to a system-level viewpoint. My second reflection is that the application of AI innovations to improve healthcare requires a multidisciplinary team to bring it to fruition, and it requires multi-stakeholder effort. It takes a village. Let me explain, using SELENA+ as an illustration. SELENA+ is the National AI project from the healthcare sector. Jointly developed by the Singapore National Eye Centre, the Singapore Eye Research Institute and the NUS School of Computing, it is an AI algorithm that analyses eye scans to detect diabetic retinopathy. SELENA+ has achieved accuracy levels of over 90%, on par with the performance of professional graders but doing it in a much shorter time.
SELENA+ is not a bolt out of the blue, but is the latest chapter of innovations in remote diagnostic imaging. Singapore’s efforts in teleradiology started at least a decade and half ago, including our x-rays being read in India, and US x-rays being read by us to take advantage of time-zone differences. We gradually learnt to be assured of clinical standards and patient safety, critical factors for successful healthcare innovations. I recall SNEC then introduced the remote diagnosis of diabetic retinopathy, for our patients in our polyclinics. The ability of our eye specialists in our national specialty centre to reach a much broader swathe of patients in the primary care system was a huge step forward, enabling us to detect and treat patients much earlier than before. Early detection was immensely valuable to tackling diabetes. Making this happen required collaboration between our specialty centres and our primary care network in many things, including workflow processes, patient management and data transfers. It required clinicians, nurses, administrators, technical specialists, and IT professionals to come together. It took hard work to resolve obstacles and build collaborations. Because we made this work manually so to speak, introducing SELENA+ now was made smoother. It takes a village, with sustained determination and commitment over time, to make innovations happen.
SELENA+ has been deployed in most polyclinics already and will reach all polyclinics by the end of 2021. It is projected to screen 120,000 patients annually, covering all diabetic eye screenings in our polyclinics across the nation. SELENA+ has also been licensed to local start-up EyRis, which has business in more than 20 countries and regions including EU, China and SEAsia.
The journey continues. We have an on-going AI in Healthcare Grand Challenge for Singapore, which aims to tackle major chronic diseases afflicting Singaporeans such as diabetes mellitus, hypertension and hyperlipidemia. I look forward to multidisciplinary teams of researchers, clinicians, computer scientists and data analysts developing more innovative solutions for this.
The importance of ethical AI and trust, and Singapore’s thought leadership
My third point to leave you with, is that our adoption of AI must be premised upon its safe and ethical use. We must manage a delicate balance between encouraging innovation while maintaining a high level of public trust in the technologies deployed. This is especially so in healthcare where misapplication can cause great harm to patients and destroy trust.
Addressing concerns about the risk of harm from using AI requires us to think deeply about questions that span policy, legal, technological and ethical domains. For example:
a. AI can be susceptible to errors and bias. How do we assess what level of accuracy is considered acceptable, and how might this vary across use cases?
b. How do we ensure that AI solutions are sufficiently robust to guard against irresponsible deployment of AI algorithms or improper management of training data? And to protect against cybersecurity risks, including data manipulation?
c. How do we determine accountability, if an AI system goes awry? In real life complications, who would be responsible?
These challenges apply to AI everywhere, but they have a particular resonance in healthcare because it affects patients’ health and their lives. So in your discussions and in developing innovations, do build in governance, ethical and resilience considerations into the design of the AI innovation.
Singapore has taken concrete measures to help establish principles and practical guidelines to ensure that AI is deployed in a responsible and ethical manner. This is critical for public and professional trust in AI. In Jan 2020, Singapore released the 2nd edition of the Model AI Governance Framework. This translates ethical principles into practical measures that organisations can adopt. In Oct this year, MOH issued to healthcare providers its AI in Healthcare Guidelines or AIHGle, with recommendations to support patient safety and improve trust in AI and sharing best practices with AI system developers and implementation teams. For AI medical device manufacturers, I understand that this complements HSA’s regulations for AI Medical Devices.
Various companies in the healthcare sector have implemented the Model AI Governance provisions. To cite one example, UCARE.AI which is a local healthcare start-up, has developed an AI-powered cost predictor to derive accurate estimates of hospital bills for patients. It has been transparent in its usage of AI, clients with concerns about bill predictions were encouraged to highlight them; and UCARE.AI also encrypted sensitive data and minimised usage of this data to preserve patient confidentiality. I understand deployment of the AI solution has gone well, with hospitals providing more accurate bill estimates, and patients having greater peace of mind over the likely size of these bills.
Singapore is also actively involved in international discussions on ethical AI. For example, IMDA and PDPC are developing an AI Governance testing framework, consistent with internationally recognized principles from EU, UNESCO, OECD and others. Best practices in AI feature in our Digital Economy Agreements. These various efforts support Singapore’s vision to become a global hub for the development, test-bedding, deployment and scaling of AI solutions.
Let me conclude by wishing you a rich and fruitful Datathon. The conversations in panels and workshops these few days bring to the fore some of the latest solutions for today’s healthcare issues. I hope that they will inspire you to develop innovations that will benefit our healthcare system and the patients that you serve.
My congratulations to NUHS, NUS and MIT Critical Data for making this event happen. All the very best to you and the participants. Thank you.
-------------------------------------------------------------------------------------------------------------------------- 1 Identifying Infection Disease Combination Therapy with AI