基于混合注意力机制的脑卒中康复方案推荐

Rehabilitation Program Recommendation for Stroke Based on Hybrid Attention Mechanism

脑卒中是一种急性脑血管疾病,患者康复周期长,康复机制复杂,给家庭和社会带来极大负担。为了弥补脑卒中传统诊疗方法的不足,为经验不足的医师提供有效的临床辅助决策支持,提出了一种面向电子病历的混合注意力机制模型为脑卒中患者的康复治疗方案进行推荐。该模型首先利用康复项目语义信息挖掘其与病历文本之间的语义关系,构建面向特定康复项目的病历表示;同时采用自注意力机制从病历文本中识别特定康复项目的病历表示;然后将二者进行有机地融合,生成最终的病历表示;最后构建多标签分类器进行康复治疗项目的 top-N推荐。在一个真实的来自三甲医院神经内科和康复医学科的电子病历数据集上进行实验,验证了模型的有效性。

Stroke is an acute cerebrovasrular disease. It has a long recovery period and a complicated recovery mechanism, which brings a great burden to the family and society. In order to make up for the shortcomings of traditional treatment methods and provide effective clinical auxiliary decision support for inexperienced doctors, this paper proposed a hybrid attention mechanism model for electronic medical records(EMRs) to recommend the rehabilitation treatment program for stroke patients. First, the model takes advantage of rehabilitation items semantic information to mine the semantic connection between rehabilitation items and medical records;meanwhile, the self-attention mechanism is used to identify the medical records representation from the medical record content information. Then, the two parts are organically fused to generate the final medical record representation;Finally, a multi-label classifier is constructed for top-N recommendation for rehabilitation items. Experiments are carried out on a real EMRs data set from the department of Neurology and rehabilitation medicine of a first class hospital to verify the effectiveness of the model.