基于自然语言技术进行血管超声报告智能生成诊断提示的探索

Investigation on Intelligent Generation of Diagnostic Suggestion in Vascular Ultrasound Reports By Natural Language Technology

目的:探索基于血管超声报告进行自然语言处理并智能生成诊断提示的方法。方法:对血管超声报告采用Bi-LSTM-CRF模型进行分词及命名实体识别,根据专家指南构建知识图谱,通过查询实体及关系属性,实现超声诊断的智能建议。结果:Bi-LSTM-CRF模型识别了4类实体,其识别准确度均在90%以上,单个报告的处理速度在毫秒量级,同时构建包括4类关系的知识图谱。结论:自然语言处理结合知识图谱可以用于血管超声报告的智能诊断提示生成。

Objective:To explore the method of intelligent generation of diagnostic tips in vascular ultrasound reports using natural language processing technology.Methods:This article takes vascular ultrasound report as the research object.Bi-LSTM-CRF model was used for word segmentation and named entity recognition,and knowledge graph was constructed according to the expert's guide.By querying the entity and relationship attributes,intelligent suggestions for ultrasound diagnosis were generated.Results:The natural language processing model recognized four types of entities,and the recognition accuracy was over 90%.The processing speed of a single report was in the order of milliseconds.At the same time,the knowledge graph of vascular ultrasound including four kinds of relationships is constructed.Conclusion:Natural language processing combined with knowledge graph can be used for intelligent diagnosis suggestion generation of vascular ultrasound report.