基于集成学习的伤情分类技术研究

Triage classification technology based on ensemble learning

目的在大规模伤员救治中,伤情分类作为伤员救治的重要组成部分,始终贯穿在救援的全过程之中。本文旨在依照伤员的实时伤情对其进行检伤分类研究,以便实施针对性救护处置,最大程度降低伤亡率与致残率。方法伤员的伤情分类通常分为4种,即轻伤、中度伤、重伤、危重伤,分别对应卫勤急救分类中的常规处置、优先处置、紧急处置和期待处置。为此,基于医院收集的急救数据,采用“一对一”与“一对多”2种多类别分类策略,结合集成学习方法与投票决策机制,实现急救伤情的准确分类。结果基于“一对一”策略的集成学习模型对4类伤情分类的平均准确率可达89%,高于基于“一对多”策略模型取得的87%的平均准确率。结论该技术能够基本实现对伤情快速有效地分类,为伤员救治提供重要参考。

Objective As a critical part of the treatment of a large number of the wounded,triage of the injured soldiers runs through the whole process of emergency rescue.The significance of triage is to treat casualties based on their current conditions,aiming to reduce the casualty and disability rate greatly.Methods The injuries are usually divided into four classes,including minor injury,moderate injury,severe injury,and critical injury,which correspond to the conventional treatment,priority treatment,emergency treatment and expected treatment in the medical first aid classification,respectively.Therefore,based on the emergency data collected by the hospital,two multi-class classification strategies,“one versus one”(OVO)and“one versus all”(OVA)were adopted in this research.Combined with ensemble learning methods and voting decision-making mechanisms,this research realized accurate classification of emergency injuries.Results The average accuracy of the ensemble learning model based on OVO strategy was 89%,which was higher than the average accuracy(87%)based on OVA strategy model.Conclusion The proposed method can realize the rapid and effective triage and provide important references for the treatment strategies of the wounded.