A new editorial suggests that the future of medical artificial intelligence lies not only in diagnosing diseases after they appear, but also in identifying subtle biological changes that occur much earlier—before any symptoms develop.
Published in Intelligent Medicine, the article by researchers from leading Chinese institutions highlights a shift toward analyzing how health data evolve over time. By integrating information from genomics, medical records, imaging, and wearable devices, AI systems could potentially detect critical ‘tipping points’ that signal the body is moving toward disease.
At the core of this approach is dynamic network biomarker (DNB) theory, which tracks sudden increases in fluctuations and correlations within biological networks. Previous studies have shown that this method can detect early warning signs of illness, such as instability in gene expression before flu symptoms appear or transitions of cells from healthy to cancerous states—with prediction accuracy exceeding 80%.
Another promising tool is individual-specific edge-network analysis (iENA), which allows clinicians to assess disease risk using data from a single patient over time, without relying on comparison groups. In gene expression studies, this method has demonstrated very high accuracy, making real-time, personalized monitoring more feasible.
The authors also emphasize the value of combining traditional physiological knowledge with modern deep learning. For example, hybrid AI models have significantly improved blood glucose predictions in type 1 diabetes, reducing errors by more than half compared with conventional approaches. These systems can even create ‘digital twins’ of patients, enabling doctors to simulate treatments before applying them in real life.
Advances are also being seen across other data types. AI models using electronic health records have improved diagnostic predictions, while brain imaging–based models can forecast treatment outcomes in neurological conditions. Transformer-based systems trained on long-term patient data are also capable of predicting risks for multiple diseases, including diabetes and hypertension.
Despite these advances, researchers stress that AI should support—not replace—clinical decision-making. These tools are designed to provide early warning signals and enable preventive care while preserving the essential role of physicians.
However, several challenges remain. Data inconsistencies and missing information can lead to false alerts. Current models are better at identifying patterns than at establishing cause-and-effect relationships. In addition, many AI systems still lack transparency, which can limit trust among clinicians.
Ethical concerns also persist, including data privacy risks and potential algorithmic bias, especially when models are applied to populations that differ from those on which they were trained.
Looking ahead, the authors identify two key priorities: integrating multiple types of medical data into unified models and conducting rigorous clinical validation through real-world studies. They conclude that only through careful testing and responsible implementation can AI fulfill its promise of transforming healthcare from reactive treatment to proactive prevention.
