Peking University and Wenzhou Medical University research team establishes generative multimodal cross-organ medical imaging infrastructure model (MINIM)

Artificial intelligence is becoming more and more important in the medical field, especially in medical image analysis, but there are barriers to obtaining high-quality and diversified medical image data, such as patient privacy protection and high cost of data labeling, for which researchers have explored synthesizing data with generative AI technology to expand it. A team of researchers from Peking University and Wenzhou Medical University has established a generative multimodal cross-organ medical image base model (MINIM), which can synthesize large amounts of high-quality medical image data based on textual commands and multiple imaging modalities across multiple organs, providing technical support for related medical work. MINIM is similar to an "image generator", which can synthesize a variety of medical image data according to textual descriptions, and its synthesized images are highly consistent with real images in many aspects. Experimental results show that the synthesized data reaches the international leading level in terms of doctors' subjective evaluations and objective testing standards, and the accuracy of multidisciplinary medical tasks can be improved by using 20 times of the synthesized data. The synthetic data generated by MINIM has a broad application prospect, and can be used alone or mixed with real data to train medical image models to improve model performance, and its synthetic data training has already shown performance improvement in key areas such as disease diagnosis.
Paper address: https://www.nature.com/articles/s41591-024-03359-y