Hello, JNPMEDI has prepared medical device clinical trial plan approved in April 2025.ย
Shall we take a closer look? ๐
ย 
ย
This month, the Ministry of Food and Drug Safety(MFDS) approved clinical trials for a wide range of indications involving various digital technologies, including Generative AI-powered chest X-ray interpretation software, as well as digital imaging diagnostics, rehabilitation, and internal function assessment.
Notably, many approvals reflect real-world clinical needs, such as robot-assisted rehabilitation devices for elderly motor recovery and autonomic nervous system analysis software. The trend is shifting toward a comprehensive evaluation that goes beyond diagnostic accuracy to include user accessibility, adherence, and clinical improvement. This signals that the paradigm for evaluating digital medical devices based on real-world use is advancing.
A key highlight this month is โAIRead-CXR,โ a software that uses Generative AI to automatically draft chest X-ray reports. Distinct from traditional analytical AI, this solution is currently undergoing a multi-center, retrospective, randomized, crossover confirmatory clinical trial.
According to the MFDSโs โGuideline on Generative AI Medical Devicesโ (released Jan 24), validation must focus on how medical staff interpret and use generated outputs in clinical settings. Beyond quantitative metrics, qualitative criteria like RADPEER scores and a 5-point scale for clinical significance are applied. This approach accounts for Generative AIโs characteristic of producing diverse text outputs for the same input.
Since performance can vary by user, it is essential to clearly define the expertise of participating medical staff and include subgroup analyses in the study design.
Regarding technical documentation, the unique characteristics of Generative AI are emphasized. The AI model architecture (GAN, VAE, LLM, etc.) and its operational flow must be clearly described, along with evaluation metrics like BERT Score, BLEU, and ROUGE to quantify the quality of the generated text. It also requires technical explanations and mitigation strategies for risks unique to Generative AI, such as hallucination and drift. Training data quality, domain suitability, and outlier handling are also critical review factors.
From a risk management perspective, the trial considers not only errors in the generated output but also risks arising from a user's misinterpretation or over-reliance on it. To address this, submissions must include measures to minimize user interface (UI) confusion, result interpretation guidelines, a cybersecurity response plan, and usability evaluation data. Beyond the clinical trial itself, establishing post-market surveillance strategies based on Real-World Evidence (RWE) is also recommended.
โAIRead-CXRโ is one of the first Generative AI cases undergoing full regulatory review and is expected to set a benchmark for future approvals. By incorporating user-centric qualitative evaluation and multi-layered risk management, this trial marks a significant turning point for the safe and effective use of Generative AI in healthcare.
We will continue to share useful insights on clinical trials.ย
Thank you! ๐
๐ References
- https://emedi.mfds.go.kr/msismext/emd/ifm/ctestAprvStatusView.do
- https://www.mfds.go.kr/brd/m_99/view.do?seq=48833&srchFr=&srchTo=&srchWord=&srchT[โฆ]&itm_seq_2=0&multi_itm_seq=0&company;_cd=&company;_nm=&page=1
Hello, JNPMEDI has prepared medical device clinical trial plan approved in April 2025.ย
Shall we take a closer look? ๐
ย
ย
This month, the Ministry of Food and Drug Safety(MFDS) approved clinical trials for a wide range of indications involving various digital technologies, including Generative AI-powered chest X-ray interpretation software, as well as digital imaging diagnostics, rehabilitation, and internal function assessment.
Notably, many approvals reflect real-world clinical needs, such as robot-assisted rehabilitation devices for elderly motor recovery and autonomic nervous system analysis software. The trend is shifting toward a comprehensive evaluation that goes beyond diagnostic accuracy to include user accessibility, adherence, and clinical improvement. This signals that the paradigm for evaluating digital medical devices based on real-world use is advancing.
A key highlight this month is โAIRead-CXR,โ a software that uses Generative AI to automatically draft chest X-ray reports. Distinct from traditional analytical AI, this solution is currently undergoing a multi-center, retrospective, randomized, crossover confirmatory clinical trial.
According to the MFDSโs โGuideline on Generative AI Medical Devicesโ (released Jan 24), validation must focus on how medical staff interpret and use generated outputs in clinical settings. Beyond quantitative metrics, qualitative criteria like RADPEER scores and a 5-point scale for clinical significance are applied. This approach accounts for Generative AIโs characteristic of producing diverse text outputs for the same input.
Since performance can vary by user, it is essential to clearly define the expertise of participating medical staff and include subgroup analyses in the study design.
Regarding technical documentation, the unique characteristics of Generative AI are emphasized. The AI model architecture (GAN, VAE, LLM, etc.) and its operational flow must be clearly described, along with evaluation metrics like BERT Score, BLEU, and ROUGE to quantify the quality of the generated text. It also requires technical explanations and mitigation strategies for risks unique to Generative AI, such as hallucination and drift. Training data quality, domain suitability, and outlier handling are also critical review factors.
From a risk management perspective, the trial considers not only errors in the generated output but also risks arising from a user's misinterpretation or over-reliance on it. To address this, submissions must include measures to minimize user interface (UI) confusion, result interpretation guidelines, a cybersecurity response plan, and usability evaluation data. Beyond the clinical trial itself, establishing post-market surveillance strategies based on Real-World Evidence (RWE) is also recommended.
โAIRead-CXRโ is one of the first Generative AI cases undergoing full regulatory review and is expected to set a benchmark for future approvals. By incorporating user-centric qualitative evaluation and multi-layered risk management, this trial marks a significant turning point for the safe and effective use of Generative AI in healthcare.
We will continue to share useful insights on clinical trials.ย
Thank you! ๐
๐ References
- https://emedi.mfds.go.kr/msismext/emd/ifm/ctestAprvStatusView.do
- https://www.mfds.go.kr/brd/m_99/view.do?seq=48833&srchFr=&srchTo=&srchWord=&srchT[โฆ]&itm_seq_2=0&multi_itm_seq=0&company;_cd=&company;_nm=&page=1