Response to Comments on “Artificial Intelligence-Driven Drafting of Chest X-Ray Reports: 2025 Position Statement From the Korean Society of Thoracic Radiology Based on an Expert Survey”

We would like to express our sincere gratitude to the Letter’s authors [1] for their interest in our article, “Artificial Intelligence-Driven Drafting of Chest X-Ray Reports: 2025 Position Statement From the Korean Society of Thoracic Radiology Based on an Expert Survey” [2] and for their thoughtful and constructive comments. We deeply appreciate their detailed review and are pleased to note their agreement with the Korean Society of Thoracic Radiology’s balanced perspective regarding the current capabilities, limitations, and future trajectory of generative artificial intelligence (AI) in chest X-ray reporting.

As the authors correctly highlighted, a central finding of our study was that the performance of the AI-based report drafting tool is highly context-dependent [2]. While the tool achieved a promising accuracy of 86.0% in health-screening settings characterized by a predominance of normal cases, its reliability significantly declined in high-acuity environments like intensive care units to between 60.0% and 69.0%. This discrepancy underscores the gap in handling complex clinical scenarios involving multifactorial findings, aligning with recent findings that domain-specific generative AI models remain susceptible to “hallucinations” or omissions in complex tasks [3]. Therefore, we strongly agree that deployment must be stratified by risk profile, prioritizing low-complexity screening while exercising extreme caution in acute care settings.

We also concur that the “neutral” consensus implies “cautious optimism,” as radiologists face pressure to adopt AI to mitigate workforce shortages and burnout [4]. However, the “human-in-the-loop” approach remains necessary to ensure safety [5], as there is a significant risk of “automation bias” where radiologists—especially those with less experience—may over-rely on erroneous AI drafts [6]. Consequently, confirmation by radiologists remains indispensable, and legal frameworks regarding liability must be clearly established to address issues where AI errors lead to patient harm [7]. Future research must now advance from subjective expert surveys to large-scale studies with objective evaluation standards.

In conclusion, we thank the Letter’s authors for their valuable contribution to this discussion. We hope that our position statement and this ongoing dialogue will serve as a foundational guide for the responsible, evidence-based adoption of AI-driven report drafting tools in radiology practice. Given the rapid evolution of AI in medicine, this statement may require timely updates to maintain its relevance. Radiological societies bear a critical responsibility to objectively evaluate safety and provide guidelines to prevent misuse or moral hazard. These efforts must be sustained, evolving in tandem with future technological progress.