A new study introduces CT-RATE, a public dataset that includes 25,692 non-contrast 3D chest CT scans paired with their corresponding radiology reports. This dataset aims to advance medical imaging AI, particularly in 3D imaging, which has previously been hindered by the lack of comprehensive datasets. Researchers have developed CT-CLIP, a contrastive language-image pretraining framework utilizing CT-RATE, which shows improved performance in multi-abnormality detection compared to fully supervised models. Additionally, a vision-language chat model, CT-CHAT, was created for interactive analysis of 3D chest CT volumes, fine-tuned on over 2.7 million question-answer pairs from the dataset. These innovations address key challenges in 3D medical imaging and are expected to enhance future medical AI applications and patient care. The dataset and related models are openly accessible for further research.
Thu, 12 Feb 2026 23:33:41 GMT | Nature