Osaka Researchers Develop AI System to Fix Radiology Labeling Errors

Artificial intelligence is becoming a powerful tool in modern healthcare, especially in radiology. Hospitals around the world now use deep-learning systems to analyze X-ray images and support doctors in diagnosis and research.

However, these AI systems depend heavily on one important step that is often overlooked: correct image labeling. Researchers at Osaka Metropolitan University have now developed an AI-based solution to fix this problem and improve the reliability of radiology data.

In most hospitals, X-ray images are labeled manually before they are stored or used to train AI models. These labels include information such as the body part being scanned, the direction of the X-ray, and whether the image is rotated or inverted. In busy clinical environments, mistakes are common. Labels may be missing, inconsistent, or simply wrong. When these flawed images are used to train deep-learning systems, the AI can learn incorrect patterns, leading to unreliable results.

The Osaka research team identified labeling errors as a major weakness in medical imaging AI. They found that even small mistakes when repeated across large databases can grow into thousands of mislabeled images. These errors reduce the accuracy of AI systems that are designed to handle routine clinical tasks and advanced research work.

To solve this issue, the researchers developed two AI models that automatically check and correct radiology labels. The first model, known as Xp-Bodypart-Checker, examines X-ray images and identifies which body part is shown. This helps detect cases where an image is labeled incorrectly, such as a chest X-ray being marked as an abdomen scan.

The second model, called CXp-Projection-Rotation-Checker, focuses on chest X-rays. It checks how the image was taken, including the projection direction and whether the image is rotated or flipped. These details are important because incorrect projection or rotation labels can confuse AI systems and lead to faulty analysis.

The results of the study were impressive. Xp-Bodypart-Checker achieved an accuracy of 98.5 percent, while CXp-Projection-Rotation-Checker also reached 98.5 percent accuracy for projection detection and 99.3 percent for rotation detection. These high scores show that AI can reliably catch labeling errors that are easy to miss during manual checks.

The researchers believe that combining both models into a single system could greatly improve the quality of radiology data used in hospitals. Cleaner, more accurate labels would lead to stronger AI performance, better research outcomes, and safer clinical use.

Looking ahead, the team plans to further refine the system by retraining it on difficult cases, including images that were incorrectly flagged or wrongly labeled. Their work highlights a simple but powerful idea: improving data quality is just as important as building smarter AI, and fixing labeling errors is a key step toward more trustworthy medical imaging systems.

With AI making its way into various technologies like medical radiology and sound technology as exemplified by the products of Datavault AI Inc. (NASDAQ: DVLT), there appears to be no industry or vertical that will be left behind by the transformative influence of AI.

About AINewsWire

AINewsWire (“AINW”) is a specialized communications platform with a focus on the latest advancements in artificial intelligence (“AI”), including the technologies, trends and trailblazers driving innovation forward. It is one of 75+ brands within the Dynamic Brand Portfolio @ IBN that delivers: (1) access to a vast network of wire solutions via InvestorWire to efficiently and effectively reach a myriad of target markets, demographics and diverse industries; (2) article and editorial syndication to 5,000+ outlets; (3) enhanced press release enhancement to ensure maximum impact; (4) social media distribution via IBN to millions of social media followers; and (5) a full array of tailored corporate communications solutions. With broad reach and a seasoned team of contributing journalists and writers, AINW is uniquely positioned to best serve private and public companies that want to reach a wide audience of investors, influencers, consumers, journalists, and the general public. By cutting through the overload of information in today’s market, AINW brings its clients unparalleled recognition and brand awareness.

AINW is where breaking news, insightful content and actionable information converge.

To receive SMS alerts from AINewsWire, text “AI” to 888-902-4192 (U.S. Mobile Phones Only)

For more information, please visit www.AINewsWire.com

Please see full terms of use and disclaimers on the AINewsWire website applicable to all content provided by AINW, wherever published or re-published: https://www.AINewsWire.com/Disclaimer

AINewsWire
Los Angeles, CA
www.AINewsWire.com
310.299.1717 Office
Editor@AINewsWire.com

AINewsWire is powered by IBN