For a patient with a severe lung infection, a single test is like a snapshot capturing the current state. However, doctors also need to predict the future course. This invention introduces an AI framework that integrates various aspects of the patient’s story. It gathers clinical indicators, lab tests, medical images, and treatment responses. Different parts analyze CT or X-ray images, medical text, and structured clinical data, while a large language model uncovers additional links among patient records, lab results, and external medical knowledge. A self-attention mechanism then connects image data, text, structured info, time-series data, and LLM-derived features, learning which relationships are crucial for the task. The goal is not just to identify the disease but to flag patients at high risk of severe pneumonia and anticipate needs like ICU care or ventilation. These predictions, combined with a language model and expert knowledge graph, provide interpretable support for treatment and resource decisions. This shifts AI from asking “What disease does this patient have?” to a more practical question: “What could happen next, and what does the clinical team need to prepare for?” The aim is not to replace clinicians but to offer a comprehensive view of the patient’s evolving condition before critical decisions are made.
Observing multiple moments in a critically ill patient
Patent number: CN 120221056 A
Inventor(s): Z. Han, X. Li, K. Cen, and Z. Liu
Citation: Z. Han, X. Li, K. Cen, and Z. Liu, “基于多模态的医学算法模型构建系统” [Multimodal-based medical algorithm model construction system], China Patent Application Publication CN 120221056 A, Jun. 27, 2025.
