行業指南

FDA對人工智慧醫療器材的監管

在美國,用於診斷、治療或提供病人照護資訊的 AI 軟體,通常由美國食品藥物管理局(FDA)作為醫療器材進行監管。

聆聽此指南AI 語音敘述 · ~3 min

本頁4 分鐘閱讀
  1. 概述
  2. 深入探討
  3. 現實世界的實施
  4. The Future of FDA Regulation of AI Medical Devices
  5. 付諸實踐
  6. 不斷探索
  7. 常見問題

概述

It reaches the market through one of three main routes: 510(k) clearance, De Novo classification, or premarket approval (PMA). This matters because it decides what evidence a developer must produce, how an AI model can be updated after launch, and how much doctors and patients can trust the tool.

深入探討

The FDA regulates software by its intended use, not by the technology inside it. If software is meant to diagnose, treat, prevent, or mitigate disease, it is generally a device; standalone software of this kind is called Software as a Medical Device (SaMD). Devices fall into Class I, II, or III by risk, and the class largely sets the pathway. Most AI-enabled devices reach the market through 510(k) clearance, in which the manufacturer shows substantial equivalence to a legally marketed "predicate" device. When a novel device is low to moderate risk and has no predicate, the De Novo pathway creates a new classification that later devices can use as a predicate. Premarket approval (PMA), the strictest route, is reserved for high-risk Class III devices and usually requires clinical evidence. The FDA's public list of authorized AI-enabled devices now runs to more than a thousand products, most of them in radiology. Not all health software is a device. The 21st Century Cures Act (2016) excludes certain clinical decision support tools. To qualify, the tool must show clinicians the basis for its recommendations so they can review them independently, and it must not analyze medical images or signals. Traditionally, a significant change to a cleared device needs a new submission. The Food and Drug Omnibus Reform Act of 2022 gave the FDA authority to approve predetermined change control plans (PCCPs), and the FDA finalized guidance for AI-enabled devices in December 2024. A PCCP lets a manufacturer make specified, pre-validated changes without filing again. A common misconception is that cleared AI devices learn on their own in the field. Most authorized models are "locked": they change only through controlled, versioned updates. In the EU, the Medical Device Regulation (MDR) classifies software under Rule 11, which puts most diagnostic software in Class IIa or higher and so requires notified body review. The EU AI Act adds high-risk AI obligations for these devices, phasing in later than most of the Act.

現實世界的實施

A radiology startup builds software that flags suspected brain bleeds on CT scans and moves them up the reading queue. It gets 510(k) clearance by showing substantial equivalence to a triage tool that is already cleared.

IDx-DR (now LumineticsCore) detects diabetic retinopathy in retinal photos without a specialist reading the image. There was no predicate device, so the FDA authorized it through the De Novo pathway in 2018.

A maker of ECG analysis software submits a predetermined change control plan (PCCP) with its application. The plan describes how the company will retrain the model on new data and what tests each update must pass, so those updates do not each need a new submission.

A company selling the same imaging AI in Europe must get a CE mark under the EU Medical Device Regulation through a notified body. It must also prepare for the EU AI Act, which treats such products as high-risk AI systems.

The Future of FDA Regulation of AI Medical Devices

The main open question is generative AI. Large language models produce open-ended output that is hard to validate the way a fixed classifier is validated. The FDA's Digital Health Advisory Committee held its first meeting on generative AI-enabled devices in November 2024. Expect more attention to monitoring real-world performance after deployment, to transparency about training data, and to greater use of PCCPs as manufacturers learn what the FDA will accept. In Europe, companies face overlapping MDR and AI Act requirements. Guidance on how the two fit together, and whether notified bodies have enough capacity, will shape how quickly AI devices reach patients there.

付諸實踐

為什麼這很重要

  • 背景與規則. 產業背景決定了人工智慧創意能否與現實接觸。
  • 品質管控. 領域約束會影響可接受的錯誤率和監督模型。
  • 配裝選擇. 成功的部署使技術能力與第一線工作流程保持一致。

留意

  • 監理要求可能會使原本強大的原型失效。
  • 歷史資料可能會編碼損害特定社區的偏見。
  • 遺留系統可能會造成整合瓶頸和隱性成本。

後續步驟

  1. 讓領域專家參與從問題框架到評估的整個過程。
  2. 在啟動前設計審計追蹤和文件。
  3. 儘早驗證合規性和安全義務。
  4. 分階段推出,並有明確的停止和回滾標準。

不斷探索

免費電子報

獲取每日 AI 簡報

每個工作日上午提供三則經過驗證的 AI 新聞,以淺顯易懂的英文撰寫。永久免費,無廣告。

每個工作日一封電子郵件。一鍵取消訂閱。我們絕不出售或分享您的地址。

測試一下自己

Take the FDA Regulation of AI Medical Devices quiz

每個答案皆有即時回饋,通過課程後可獲得具備可驗證 ID 的可分享證書。

開始測驗

支持免費的 AI 教育。AI Understanding 是一個 501(c)(3) 非營利組織——絕無廣告,絕無付費牆。 捐款

常見問題

What is FDA Regulation of AI Medical Devices?

In the United States, AI software that diagnoses, treats, or informs care for a patient is usually regulated by the FDA as a medical device. It reaches the market through one of three main routes: 510(k) clearance, De Novo classification, or premarket approval (PMA). This matters because it decides what evidence a developer must produce, how an AI model can be updated after launch, and how much doctors and patients can trust the tool.

What must a manufacturer show to obtain 510(k) clearance?

A 510(k) rests on substantial equivalence to an existing predicate. Most AI-enabled devices reach the market this way.

The FDA authorized IDx-DR, an autonomous AI for diabetic retinopathy, in 2018. Which pathway did it use, and why?

De Novo is for novel low-to-moderate-risk devices with no predicate. Once a device is authorized this way, later devices can use it as a predicate.

What does a predetermined change control plan (PCCP) allow?

A PCCP authorizes planned changes, such as retraining, in advance, along with the protocol used to validate them. Changes outside the plan still need a new submission.

Which three parts make up a PCCP?

A PCCP describes the planned changes, sets out how they will be developed and validated, and assesses their benefits and risks.

Which statement about most FDA-authorized AI devices is accurate?

Most authorized AI models are locked. Updates come as controlled versions, and radiology, not dermatology, accounts for most authorized AI devices.