Public dataset aims to teach AI to 'show its work' on CT scans
Earlier this summer, an international team led by biomedical informatics researchers at Harvard Medical School released the first public dataset that connects descriptive radiological information, such as "3 mm nodule in the lower left lobe," to precise locations in chest CT scans.
Earlier this summer, an international team, led by researchers at Harvard Medical School, unveiled the first public dataset that links radiology report descriptions to precise locations within chest CT scans. This new dataset, named ReXGroundingCT, was created by biomedical informatics experts to help AI models generate accurate and detailed reports, paired with visual markers that show clinicians precisely where to look across the complex 3D images.
The main goal of ReXGroundingCT is to make radiology findings clearer and easier to verify, turning AI-generated descriptions into evidence clinicians can confirm. The dataset, released in NEJM AI, consists of over 3,142 chest CT scans, with more than 16,000 manually annotated abnormalities each double-checked by a board-certified radiologist.
The researchers hope ReXGroundingCT will serve as a benchmark for training AI models that can not only describe findings but also show exactly where they exist within the scans. This tool could improve patient experience by generating video explanations of radiology findings and help train medical students and radiologists in report writing and localization of anomalies.
Written by urgent.news from Medical Xpress's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
