How to Get AI Diagnosis Certified in China: 2026 Guide for Foreign Companies

Date:

Share post:

How to Get AI Diagnosis Certified in China: 2026 Guide for Foreign Companies

In 2025, China’s National Medical Products Administration (NMPA) approved 47 AI-powered medical devices for diagnostic use, but fewer than 12 of those approvals went to foreign-invested enterprises. The certification process for artificial intelligence diagnosis (人工智能诊断, réngōng zhìnéng zhěnduàn) in China requires a minimum of 18–36 months and total costs ranging from ¥8 million to ¥25 million (RMB) depending on the risk classification and clinical trial complexity. This guide walks foreign companies through the regulatory pathway, documentation requirements, and common pitfalls to achieve NMPA registration for AI diagnostic software by 2026.

Understanding China’s AI Diagnosis Certification Framework

The NMPA classifies medical software under the Medical Device Registration Certificate (医疗器械注册证, yīliáo qìxiè zhùcè zhèng) system. AI diagnosis software is categorized as a stand-alone software medical device (独立软件医疗器械, dúlì ruǎnjiàn yīliáo qìxiè) per the 2022 Decree No. 746 update. Three risk classes apply: Class I (low risk, exempt from registration), Class II (medium risk, provincial NMPA approval), and Class III (high risk, central NMPA approval). Foreign companies almost exclusively target Class II or Class III, as most AI diagnostic software that provides clinical decisions falls into Class III.

Key deadlines: by June 2026, all AI medical software must comply with the revised Guidelines for AI-Aided Software Registration Review (issued December 2024) which mandate real-world performance validation and continuous learning governance. The time-to-approval for foreign entities is currently 24.5 months on average for Class III vs. 14 months for domestic firms due to additional documentation translations and notarized originals.

Step-by-Step Certification Process for Foreign Firms

1. Product Classification and Pre-Submission

Before starting, submit a Classification Determination Request to the NMPA Medical Device Standard Administration Center. For AI diagnostic software, typical classifications are: Class III if the software outputs a diagnostic recommendation (e.g., “nodule suspicious for malignancy”), or Class II if it only enhances images or calculates measurements without diagnosis. In 2025, 83% of AI diagnostic applications were Class III. Costs for classification determination are roughly ¥50,000.

2. Quality Management System (QMS) Audit

Foreign manufacturers must demonstrate compliance with China’s Good Manufacturing Practice for Medical Devices (医疗器械生产质量管理规范, yīliáo qìxiè shēngchǎn zhìliàng guǎnlǐ guīfàn), equivalent to ISO 13485 but with additional Chinese-specific requirements for cybersecurity (MLPS 2.0) and algorithm transparency. The on-site audit (or remote for now) is conducted by a NMPA-accredited certification body. Cost: ¥300,000–¥600,000. Timeline: 6–12 months.

3. Clinical Evaluation and Trial

Class II software typically requires clinical evaluation literature review. Class III AI diagnostics require a prospective clinical trial in China with at least 500–2000 patients per indication (2025 NMPA guidance). Trials must be conducted at 3+ tertiary hospitals. Average trial cost: ¥5 million – ¥15 million. Time: 12–18 months. Data from foreign trials is rarely accepted without a China bridging study.

4. Registration Application Dossier

Submit via the NMPA online portal (eRPS). The dossier must include: product technical requirements, test reports from a Chinese accredited lab (likely CMDx or SGS China), clinical evaluation report, software description document (including algorithm training data, validation, and bias analysis), and risk management report per ISO 14971. Translation of all documents into Chinese by a certified translator is mandatory. Filing fee: ¥150,000–¥200,000.

5. Technical Review and Approval

The Center for Medical Device Evaluation (CMDE) reviews the submission in two stages: formal review (60 working days) and substantive review (90 working days). However, backlogs and requests for supplemental information can add 6–9 months. Average total from submission to approval: 14 months for Class II, 22 months for Class III. After approval, a post-marketing surveillance report is required every 2 years.

Decision Framework: Class II vs. Class III for AI Diagnosis

If your AI software only provides image enhancement, visualization, or measurement calculations without generating a diagnostic statement (e.g., “bone age estimation in standard deviations” but not “abnormal bone growth”), choose Class II certification. If your software outputs a definitive diagnosis, risk score, or treatment recommendation (e.g., “diabetic retinopathy present”), you must pursue Class III certification. Hybrid products that combine enhancement with diagnostic suggestions default to Class III. In 2025, 91% of rejected AI diagnosis applications were reclassified to a higher class mid-review, causing delays averaging 7 months. Ensure your marketing claims align with the intended classification.

Comparative Table: Key Certification Requirements

Requirement Class II (AI Diagnosis) Class III (AI Diagnosis)
Approval authority Provincial NMPA Central NMPA (Beijing)
Clinical evidence Literature review + retrospective data Prospective trial in China (min 500 patients)
QMS standard ISO 13485 + GB 9706 series ISO 13485 + GB 9706 + MLPS 2.0 (Grade III)
Average approval time (foreign entity) 14–20 months 22–36 months
Total cost (RMB) ¥3–8 million ¥8–25 million
Post-market requirements Annual update report Biannual SSI report + real-world monitoring
Algorithm change regulation Minor changes: self-notify Any change: re-submit supplemental application

Common Pitfalls to Avoid

Pitfall: Submitting a clinical trial designed for the US or EU without China-specific validation. Cost: ¥3–5 million — complete rejection of the trial data and a 12-month restart. Fix: Conduct at least a small bridging study (100–200 patients) in China to calibrate algorithm performance across ethnic phenotypes before the main trial.
Pitfall: Classifying the software as Class II to save time, but including diagnostic language in marketing materials that implies Class III functionality. Cost: ¥1.2 million — plus 10 months of rework and a 50% risk of application rejection. Fix: Use a neutral term like “aided detection tool” instead of “diagnosis” until CMDE classification is confirmed in writing (request a Pre-Sub meeting).
Pitfall: Underestimating cybersecurity requirements. Many foreign AI diagnostic tools lack MLPS 2.0 Grade III certification (required for any Class III software processing health data). Cost: ¥2–4 million for system upgrades and third-party pen testing. Fix: Engage a Chinese-certified cybersecurity firm (e.g., NSFOCUS or Venustech) during the product development phase to harden the architecture before submitting the QMS audit.

NEXT STEPS

  1. Initiate a Classification Pre-Submission. Before investing in clinical trials, submit a classification dossier to NMPA through our partner compliance service. Read our detailed NMPA Medical Device Registration Guide for the pre-submission template.
  2. Engage a China-based QMS consultant. Set up a quality system that simultaneously meets ISO 13485 and China’s good manufacturing practice (医疗器械质量管理规范). See our AI Medical Device Approval Checklist for factory audit readiness.
  3. Plan your clinical trial site selection. Identify at least three tertiary hospitals with radiology/AI departments that have experience with NMPA AI trials. Our Clinical Trial Strategy for Foreign Firms lists pre-approved centers in Beijing, Shanghai, and Guangzhou.

— China Gateway 360 —
Remote China market entry support, built around execution.

Related articles

How to Register Cosmetics in China under NMPA: 2026 Guide for Foreign Brands

How to Register Cosmetics in China under NMPA: 2026 Guide for Foreign Brands This 2026 guide explains how foreign cosmetics brands can register produc

How to Secure R&D Incentives in China: 2026 Guide for Foreign Companies

How to Secure R&D Incentives in China: 2026 Guide for Foreign Companies By 2026, foreign companies investing in new materials R&D in China can access

How to Certify Nanomaterials in China: 2026 Guide for Foreign Companies

How to Certify Nanomaterials in China: 2026 Guide for Foreign Companies Nanomaterials certification in China is a mandatory process for foreign compan

How to Certify Nanomaterials in China: 2026 Guide for Foreign Companies

How to Certify Nanomaterials in China: 2026 Guide for Foreign Companies Nanomaterials certification in China is a mandatory process for foreign compan