‘Deepnoid’s M4CXR outperforms ChatGPT in chest X-ray’

Deepnoid said Thursday that a study titled “Comparison of Chest X-ray Interpretation Performance between Deepnoid's M4CXR and OpenAI's ChatGPT,” conducted in collaboration with Professor Lee Ro-woon of Inha University Hospital's Department of Radiology, has been published in the Journal of Clinical Medicine, an internationally recognized SCI-ranked journal.

Deepnoid's M4CXR is a cloud-based medical AI solution designed with a large language model (LLM) that analyzes chest X-ray images to generate draft readings. The study demonstrated that M4CXR is suitable as an assistant tool to empower clinicians in real-world medical settings when diagnosing chest X-rays.

Professor Lee, an expert in medical AI, led the study. When the readings of M4CXR and ChatGPT were evaluated in terms of five metrics—accuracy, false positives, location errors, count errors, and hallucinations—M4CXR consistently outperformed ChatGPT in all metrics, especially in location accuracy (76-77.5 percent), which significantly outperformed ChatGPT (36-36.5 percent). Regarding accuracy, M4CXR (60-62 percent) also outperformed ChatGPT (42-45 percent).

The study's results showed that M4CXR is suitable for use in specialized medical settings and outperformed ChatGPT, a general-purpose AI.

“This study confirms the potential of M4CXR as a diagnostic aid in the medical field,” Professor Lee said. “In the future, we plan to further strengthen the clinical validity of M4CXR by comparing its readings with those of real doctors, which will further improve the accuracy and efficiency of diagnosis.”