In the evolutionary journey of precision medicine, integrating scattered and heterogeneous medical data to enhance prediction accuracy has consistently remained a prominent research focus in data science.

Recently, Professor Vincent S. Tseng, a Chair Professor in the Department of Computer Science at National Yang Ming Chiao Tung University (NYCU), and his research team developed a pioneering "Deep Heterogeneous Multimodal Learning Technology" for predicting readmission and mortality risks in patients with heart failure. This technology achieves deep fusion of multi-source data at the technical level and outperforms traditional medical approaches in clinical validation. It was consecutively honored with the 2025 Future Tech Award and the 32nd TECO Award (Electrical Engineering/Computer Science/Communications Technology Category), setting a new benchmark for smart healthcare research in Taiwan.

Technical Core: Overcoming the "Heterogeneity" Challenge in Medical Data

The foundation of this research stems from Professor Tseng’s more than two decades of deep cultivation in data mining and Big Data analytics. As an IEEE Fellow, Professor Tseng's early contributions to high-utility mining and high-efficiency algorithms provided a solid underlying logic for processing complex, massive medical data.

In this heart failure study, the team did not simply apply existing medical AI models. Instead, addressing the fragmented nature and mismatched modalities of clinical data, they proposed an entirely new deep learning architecture. The core of this architecture lies in the deep integration of structured electronic health records, longitudinal electrocardiograms (EKG), and unstructured chest X-ray images. This approach overcomes the previous academic bottleneck where single-source data lacked sufficient accuracy when predicting complex complications.

Research Contribution: From Theoretical Algorithms to Clinical Validation—A Major Breakthrough in Heart Failure Risk Prediction

Heart failure is one of the leading causes of hospitalization among the elderly in aging societies, carrying an exceptionally high rate of readmission and risk of mortality. Developed in collaboration with Dr. Hao-Min Cheng’s team at Taipei Veterans General Hospital, Professor Tseng’s latest technology aims to break through the limitations of traditional medical forecasting.

At the technical level, the study successfully resolved the challenge of inconsistent representations among heterogeneous data. The team developed a heterogeneous representation fusion technique that precisely extracts deep features from different clinical modalities, while utilizing a dynamic weight adjustment strategy to automatically optimize the contribution of each information source during the model training process. This mechanism ensures that the model maintains exceptionally high stability when confronting real-world clinical scenarios characterized by missing information or imbalanced data. Research results demonstrate that this ensemble-based heterogeneous classifier stacking technology achieves an Area Under the Curve (AUC) of over 0.8 in predicting both short-term and long-term readmission and mortality risks, significantly outperforming internationally recognized traditional risk prediction models such as ADHERE or GWTG.

The study also underscores the scientific value of translating these research and development achievements into practical industrial applications. Currently undergoing validation across multiple medical centers, this technology effectively assists physicians in making prognostic decisions and accurately screening high-risk patient populations, thereby enhancing the quality of chronic disease care and reducing healthcare costs. This represents not only a demonstration of technological capability, but also a concrete realization of Professor Tseng’s long-term commitment to driving the integration of medicine and engineering, alongside the transformation of digital medicine.

Future Outlook: Safeguarding Every Heartbeat with Technology

Looking ahead, Professor Tseng and his team will continue to integrate Generative AI and Large Language Models (LLMs) to develop highly explainable, intelligent early-warning technologies. The ultimate goal is to make AI act like an earthquake early warning system—issuing alerts before a crisis occurs to buy critical golden time for patients.

Furthermore, these forward-looking technologies will not only be applied to smart healthcare fields such as cardiovascular diseases and dementia but will also extend to diverse arenas like environmental monitoring and smart manufacturing in the future. By combining government resources and industry-academia alliances, Professor Tseng aims to connect with top international academic and research institutions to establish large-scale international collaborative projects and innovation platforms. This initiative is not merely about R&D; it aims to construct a highly productive and creative ecosystem, thereby strengthening Taiwan's global influence in smart technology and truly transforming technological R&D into a core driving force that propels societal development and safeguards human health.