Automatic Summarization of Burn Knowledge By Systematic Review Articles

Background: In the age of information explosion, the need of artificial intelligent tool for management of huge amount of medical literature rose. We developed an unbiased and robust algorithm for automatic summarization of a knowledge domain simply by the MeSH keywords of Medline literature, and validated the feasibility in all systematic review articles of burn injury.

Materials and Methods: 492 articles were retrieved from PubMed e-Utility search interface, by keywords of "systematic"[sb] and "burns"[Major], which means all the systematic reviews with MeSH term of "burns" as major topic. The top five journals and article numbers were Burns (116), J Burn Care Res (86), Cochrane Database Syst Rev (23), J Burn Care Rehabil (11),
and Ann Plast Surg (7), from 1962 to 2017. 94% of the articles are after 2006.

All MeSH terms from the 492 Medline records were analyzed, and were clustered by overlapped terms between the articles. Two naive criteria were used for the drawing of final knowlege maps -- 1) the minimum number of intersected terms (min), and 2) the Jaccard Index (J.I.), the number of intersected terms divided by the number of union of that two articles.

Results: Five and four main knowledge skeletons were calculated without additional domain knowldege, by two criteria set [min. 4 and J.I. of 0.35] and [min. 5 and J.I. of 0.30]. The major skeleton compromised consistently of topics about epidemiology and risk analysis, which occupied about 2/3 of articles. The difference betwen the two graphs, and the small skeletons would be discussed from a surgical perspective, including the clinical practice guideline group, the wound care group, and the wound dressing group.

Conclusion: Automatic summarization of domain knowldeg and problem elucidation by MeSH terms from the free PubMed service is feasible. A large epidemiology knowledge skeleton and small wound skeletons were consistency by different criteria of the cluster analysis.