AI Medical Scans Raise Concerns: Media's Role in 'Responsible Information' is Crucial

The Hankyoreh reports on a critical symposium concerning the burgeoning role of Artificial Intelligence (AI) in South Korea's healthcare system, specifically within health check-ups. The article, titled 'AI Scans from Head to Toe... The Media's Role in Delivering 'Responsible Information' is Crucial,' underscores both the advancements and the significant challenges presented by this technological wave.

“

AI collaboration with medical professionals will be a win-win model that prevents burnout of medical staff and provides more precise and human-centered medical care to examinees.

The symposium, organized by the Korea Medical & Bio-Journalist Association, brought together a diverse group of stakeholders—medical professionals, policymakers, journalists, consumer groups, and AI startups. Presentations detailed how AI is already analyzing endoscopic images, detecting abnormalities in chest X-rays, predicting heart failure from ECGs, and assessing cardiovascular disease risk from retinal scans. While proponents like Dr. Ahn Ji-hyun of KMI highlight AI's potential to prevent physician burnout and offer more precise, human-centered care, the article pivots to a more critical perspective.

“

We must pay attention to the 'distribution shift' phenomenon where performance significantly drops when hospitals, equipment brands, and patient groups change.

Professor Kim Hyung-jin of Samsung Medical Center delivered a stark warning, dissecting four key misconceptions surrounding AI in health check-ups. He challenged the commonly cited 'over 90% accuracy rates,' revealing that real-world diagnostic accuracy can be as low as 51.1% due to 'distribution shift'—performance degradation when AI encounters data from different hospitals, equipment, or patient populations. He also differentiated between AI's ability to 'detect' anomalies and its capacity for definitive 'diagnosis,' cautioning against patient anxiety caused by AI-flagged findings that require further, often unnecessary, tests. The article also touches upon AI's tendency to learn and perpetuate existing biases, citing examples where AI models misclassified patients based on healthcare spending or exhibited gender bias in recruitment.

“

AI does not create bias; it learns bias.

Critically, the article emphasizes the media's role, as suggested by the title and the presentation by journalist Lee Ji-hyun. She observed that media coverage in South Korea has largely focused on promotional aspects—new service launches and MOUs—rather than critically examining the technical validation, ethical dilemmas, or real-world clinical challenges. This one-sided reporting, The Hankyoreh argues, can lead to an overestimation of AI's current capabilities and a neglect of crucial issues like data fragmentation, which hinders the measurement of AI's true impact. The call for 'responsible information' delivery is paramount, urging a shift towards in-depth reporting that balances innovation with rigorous scrutiny, ensuring that the public is not misled by the hype surrounding AI in healthcare.

“

When we only think about whether the technology is possible, we must also ask the question, 'Is it okay to do it?'