In Beijing, within the School of Software at Tsinghua University, the team led by Associate Professor Xu Feng has presented a development that may become a turning point in the evolution of medical artificial intelligence. The new model, called MedMPT (Medical Multimodal Pretrained Transformer), is the first foundational multimodal architecture capable of integrating visual, textual, and clinical data to support the diagnosis and treatment of respiratory diseases.
At the core of the model lies large-scale self-supervised training on a dataset of 154,274 chest CT scans and corresponding radiology reports, enabling it to merge heterogeneous data from multiple sources: images, textual conclusions, and laboratory indicators. Unlike narrow AI systems that handle a single task (such as image classification or report generation), MedMPT understands the full clinical context — from initial diagnostics to treatment recommendations.
In practical tests, MedMPT demonstrated record accuracy in detecting lung diseases, generating radiology reports, and formulating therapeutic recommendations. Its architecture allows it to work with different types of data in a single pipeline, forming the basis for an intelligent medical decision-making cycle — from patient to physician and back.
The medical industry has long needed such a universal system: respiratory diseases, including pneumonia, COPD, and post-COVID complications, require complex analysis of diverse data, making human diagnosis increasingly challenging. MedMPT offers a technological answer to this problem — a model capable of interpreting medical data like a physician and reasoning like a next-generation analytical system.
The researchers emphasize that MedMPT is not just an algorithm but a new paradigm of medical AI, where multimodal data becomes the foundation for genuine clinical reasoning by machines. This brings medicine closer to a future in which artificial intelligence becomes not merely a tool but a partner to the physician — understandable, explainable, and clinically meaningful.
Source: Tsinghua News Network, November 11, 2025.
