Duan Huilong’s team from the School of Biomedical Engineering of Hainan University published a paper in Information Sciences to achieve fast and optimal medical compliance checking

Recently, Huilong Duan's team from Hainan University collaborated with Eindhoven University of Technology in the Netherlands reported an algorithm for optimal "alignment"computation based on the coordinate plane. Dr. Hui Yan from the School of Biomedical Engineering of Hainan University is the first author, and Associate Professor Shan Nan is the corresponding author. This paper was funded by the Philips Brain Bridge project in the Netherlands. In this paper, this team addressed the long-term challenge of achieving both "efficiency" and "optimality" in the field of medical process mining. By transforming the computation problem of optimal matching into the shortest-path problem on the coordinate plane, a rigorous theoretical solution as well as practical implementation are realized. In other words, the algorithm achieving both the efficiency and optimality is proposed. The research results were published in "Information Sciences" (IF=8.233) under the title "Computing alignments with maximum synchronous moves via replay in coordinate planes".

The field of medical process mining is an important branch of medical informatics. Compliance checking aims to compare the data in real-life with standardized careprocesses to find deviations and their causes. They can thereby be used to improve the quality of care. For a long time, compliance checking technology has faced the challenge of being difficult to achieve both "efficiency" and "optimality". This is due to the optimal alignment problem is an NP-hard problem. Duan Huilong's team from Hainan University takes the real-life data (with timestamps) as the horizontal axis on the coordinate plane, and the standardized careprocess as the vertical axis of the coordinate plane (Figure 1.(a)). When an activity in the care process matches with that in real-life data, a move in the diagonal direction is made. Otherwise, a deviation represented by one move in either horizontal or vertical direction is made. A path from the origin to the end point is an alignment. On the one hand, the optimal alignment should have the largest number of "moves along the diagonal"; on the other hand, such optimal alignments is computationally expensive. Considering that the shortest path problem already has solutions that are both efficient and optimal (such as Dijkstra's algorithm), converting the optimal alignment into the problem of finding the shortest path can provide solutions for the efficient and optimal alignment computation.

Figure 1. (a) The algorithm that searches for optimal alignments using coordinate planes.

(b) A real-life care pathway model in BPMN.

(c) The number of matches in the optimal alignments between the care pathway model and real-life data.

This algorithm not only solves the problem of efficient computation of optimal alignment, but also breaks the long-term dilemma of compliance detection technology being limited to one modeling language. This method provides the opportunity to do compliance check for almost all the process languages that have execution semantics. For example, the BPMN language has been widely used in the medical field. This widely applicable and efficient algorithm proposed is able to answer "who" and "where" and "what deviation" occurred in real-world data. Moreover, these information can be used to answer"why"to facilitate decision making in a wide range of applications.

The article:

https://www.sciencedirect.com/science/article/pii/S0020025522004091?via%3Dihub