融合文本分类算法的皮肤病辅助诊疗模型  

Skin disease diagnosis and treatment model based on text classification algorithm

作　　者：凌天 诸佳珍 焦阳[1] 李露芳[1] LING Tian;ZHU Jiazhen;JIAO Yang;LI Lufang(Library of Zhejiang Chinese Medical University,Hangzhou 310053,China)

机构地区：[1]浙江中医药大学图书馆,浙江杭州310053

出　　处：《中国医学物理学杂志》2024年第8期1046-1052,共7页Chinese Journal of Medical Physics

基　　金：教育部产学合作协同育人项目(202102242027,202102242043)。

摘　　要：针对当前皮肤病辅助诊断中生物医学特征建模规模较小且耗费巨大人工成本,而患者疾病特征的时间序列同样无法准确描述等难点,本研究运用融合文本分类算法,融合常用的文本分类模型TextLSTM、TextCNN、RCNN得到皮肤疾病辅助诊疗模型(TLNN模型),通过提取图像传感器医学特征向量化后进行预处理减少焦块数量以及消除偏差较大的特征信息,提高决策数据精度。在ISIC2018和PH2数据集进行对照实验,TLNN模型的准确率为72.36%,高于其余3种文本分类模型。在与医生主观诊断对比实验中,模型诊断准确率为92%,接近于医生94%的平均准确率,而有效诊断效率(1.17min/例)明显高于医生人工诊断(4.57min/例),整体效率提升幅度达290%,结果表明对比传统人工诊断,融合文本分类算法模型能以更短时间获得精确的诊断。TLNN模型可以应用于疾病诊断,辅助医生医疗决策,为患者提供优质便捷的智能诊疗服务。In response to the challenges of small scale and huge labor cost in biomedical feature modeling in current skin disease assisted diagnosis,as well as the inability to accurately describe the time series of patient disease features,a fusion text classification algorithm is used to integrate commonly used text classification models(TextLSTM,TextCNN,and RCNN)to obtain a model based on transfer learning and neural networks(TLNN model).By extracting the medical features of image sensors and quantizing them,the pretreatment reduces the number of foci and eliminates the feature information with large deviations,thus improving the accuracy of decision data.TLNN model achieves an accuracy of 72.36%on ISIC2018 and PH2 datasets,which is higher than those of the other 3 text classification models.The diagnostic accuracy of TLNN model is close to doctor's diagnosis(92%vs 94%),but the effective diagnostic efficiency is significantly higher than doctor's diagnosis(1.17 min/case vs 4.57 min/case),and the overall efficiency is improved by 290%.The results demonstrate that the fusion text classification algorithm model can obtain accurate diagnosis in less time than the traditional manual diagnosis.TLNN model can be applied to disease diagnosis,and assist doctors in medical decision-making,thereby providing patients with high-quality and convenient intelligent diagnosis and treatment services.

关 键 词：皮肤病 辅助诊断 融合文本分类算法 D-S证据理论 医学特征 

分 类 号：R318[医药卫生—生物医学工程] TP520.60[医药卫生—基础医学]