A Decision Fusion Framework for Treatment Recommendation Systems.

Journal: Studies in health technology and informatics
Published Date:

Abstract

Treatment recommendation is a nontrivial task--it requires not only domain knowledge from evidence-based medicine, but also data insights from descriptive, predictive and prescriptive analysis. A single treatment recommendation system is usually trained or modeled with a limited (size or quality) source. This paper proposes a decision fusion framework, combining both knowledge-driven and data-driven decision engines for treatment recommendation. End users (e.g. using the clinician workstation or mobile apps) could have a comprehensive view of various engines' opinions, as well as the final decision after fusion. For implementation, we leverage several well-known fusion algorithms, such as decision templates and meta classifiers (of logistic and SVM, etc.). Using an outcome-driven evaluation metric, we compare the fusion engine with base engines, and our experimental results show that decision fusion is a promising way towards a more valuable treatment recommendation.

Authors

  • Jing Mei
    Ping An Technology, Shenzhen, China.
  • Haifeng Liu
    IBM Research China, Beijing, China.
  • Xiang Li
    Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States.
  • Guotong Xie
    Ping An Health Technology, Beijing, China.
  • Yiqin Yu
    IBM Research, Beijing, China.