AI Regulation Tracker  /  Regulator guidance

Taiwan's TFDA revises registration rules for AI diagnostic devices, effective August 14, 2025

Taiwan's Food and Drug Administration issued a revised technical guideline for AI/ML computer-aided detection and diagnosis devices, resetting what device makers and hospitals must validate before these imaging tools reach the clinic.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

Taiwan's Food and Drug Administration has revised the rulebook for the AI that reads medical images. A revised technical guideline governing the registration of AI and machine-learning computer-aided detection (CADe) and computer-aided diagnosis (CADx) medical devices took effect on August 14, 2025. It resets what manufacturers and hospitals must prove before these tools reach a clinician's workstation.

The TFDA, which sits under Taiwan's Ministry of Health and Welfare, framed the guideline around two device categories that regulators worldwide treat as distinct. CADe systems flag suspected findings, a nodule on a chest scan or a lesion on a mammogram, and leave interpretation to the physician. CADx systems go further and characterize a finding, offering a probability that it is benign or malignant. The line matters because a device that suggests a diagnosis carries different clinical risk than one that only points, and the guideline treats the two accordingly.

What the guideline requires

The core shift is standardization. The revised guideline sets defined registration and review criteria for AI/ML CADe and CADx devices, so a submission is measured against published expectations rather than assembled case by case. Alongside it, the TFDA issued an independent performance-evaluation FAQ for AI/ML devices. Together they establish that a device maker must meet standardized registration criteria and an independent performance evaluation before the software is used clinically. For radiology and pathology teams, that inserts a defined validation gate ahead of deployment.

The submissions run through the AIMD platform operated by the Smart Medical Device Project Office, which the TFDA established on May 7, 2021. That office is the administrative home for AI medical-device review in Taiwan, and the platform is where the guideline and the FAQ live for applicants to consult. Centralizing the process on one platform matters for firms that previously tracked scattered requirements, because it puts the current criteria and the evaluation expectations in a single place a regulatory team can point to.

What it does not do

The guideline is regulatory guidance, not a new statute, and it does not turn every hospital algorithm into a registered device overnight. It governs the registration and review pathway for AI/ML CADe and CADx products and the evidence behind them. It does not, on its own terms, rewrite Taiwan's broader medical-device law, and it does not remove the physician from the loop. CADe and CADx tools remain aids to a clinician, not replacements for one. Reading the guideline as a blanket approval, or as a ban, would overstate it in both directions. What it does is make the evidentiary bar explicit and consistent.

Why the workflow changes

Before a standardized criterion set, an imaging-AI vendor could face uncertainty about exactly what data, what performance thresholds, and what independent testing a reviewer would demand. The revised guideline narrows that uncertainty and pairs it with an independent performance evaluation, which shifts validation from a vendor's internal claim toward an external check. Hospitals evaluating whether to adopt a CADe or CADx tool now have a clearer question to ask a vendor: does the device meet the TFDA guideline and has it cleared the independent evaluation. That is a practical change to the buying and onboarding workflow for radiology and pathology AI, not only to the paperwork. It also gives clinical governance committees a defined artifact to file, which shortens the argument over whether a given tool has cleared an objective bar before it touches patient images. For a busy radiology department weighing several competing products, that comparability is worth as much as the criteria themselves.

The cross-border read

For a US professional, Taiwan's move is a clean comparison rather than a binding obligation. It does not reach US firms that are not registering devices in Taiwan. But it lands in the same current as the US FDA's evolving approach to software as a medical device and its Predetermined Change Control Plan framework, which addresses how a learning algorithm may update after clearance, and the European Union's Medical Device Regulation provisions that capture AI. Taiwan's decision to publish standardized CADe/CADx criteria plus an independent evaluation requirement gives regulators and buyers across the region a concrete reference point for how diagnostic-AI validation is tightening. For a US device maker with Asian ambitions, it is one more market where the imaging-AI dossier must answer a defined, published standard.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.