The reporter learned from the School of Future Technology at Peking University that a research team from Peking University and Wenzhou Medical University has established a generative multimodal cross organ medical imaging basic model (MINIM), which can synthesize massive high-quality medical imaging data based on text instructions and multiple imaging methods of multiple organs, providing strong technical support for the training of medical imaging models, precision medicine, and personalized diagnosis and treatment. This achievement has recently been published online in the internationally authoritative journal Nature Medicine. The medical imaging big model is an AI universal model trained using deep learning and large-scale data, which can automatically analyze medical images to assist in diagnosis and treatment planning. But to improve the performance of large models, a large amount of data is needed for continuous training. However, due to various factors such as patient privacy protection and high data annotation costs, there are often obstacles to obtaining high-quality and diverse medical imaging data. Therefore, in recent years, researchers have begun to explore the use of generative AI technology to synthesize medical imaging data in order to expand the data. At present, the publicly available medical imaging data is very limited, and the generative model we have established is expected to solve the problem of insufficient training data Wang Jinzhuo, an assistant researcher at the School of Future Technology at Peking University, said that the research team used high-quality image text pairing data from various organs under different imaging methods such as CT, X-ray, and magnetic resonance imaging for training, and ultimately generated a massive amount of medical composite images, which are highly consistent with real medical images in terms of image features, detail presentation, and other aspects. The experimental results show that the synthesized data generated by MINIM has reached the international leading level in subjective evaluation indicators and multiple objective testing standards for doctors, and has important reference value in clinical applications. On the basis of real data, using 20 fold synthetic data can improve the accuracy of multiple medical tasks in ophthalmology, thoracic medicine, neurology, and breast medicine by an average of 12% to 17%. Wang Jinzhuo stated that the synthesized data generated by MINIM has broad application prospects. It can be used as a separate training set to construct large medical imaging models, or combined with real data to improve the performance of the model in practical tasks and promote the wider application of AI in the fields of medicine and health. At present, using MINIM synthetic data for training in key areas such as disease diagnosis, medical report generation, and self supervised learning has shown significant performance improvements. (New Society)