基于TAO主观诊断偏差性的分类应用  

Application of classification based on TAO subjective diagnostic bias

作　　者：周亚琪 付潇 郑文 ZHOU Ya-qi;FU Xiao;ZHENG Wen(Institute of Public Safety Big Data,College of Data Science,Taiyuan University of Technology,Taiyuan 030060,China;China Institute of Marine Technology and Economy,Ship Culture Industry Development Research Center,Beijing 100081,China)

机构地区：[1]太原理工大学大数据学院公共安全大数据研究所,山西太原030060 [2]中国船舶集团有限公司综合技术经济研究院舰船文化产业发展研究中心,北京100081

出　　处：《计算机工程与设计》2024年第9期2828-2834,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(11702289);国家重点研发计划“科技助力经济2020专项”基金项目(RZ2000003782);山西省关键核心技术和共性技术研发攻关专项基金项目(2020XXX013)。

摘　　要：甲状腺相关眼病是一种早期病理特征不明显的常见慢性致盲性眼眶疾病,为解决诊断过程中医生主观诊断偏差导致的误诊问题,提出一种基于复合模型缩放方法的模型Efficient-TAO。通过优化训练过程,改进动态激活函数和随机权值平均机制提高性能,通过图像预处理算法实现眼肌CT图像数据的直接训练。实验结果表明,模型的分类准确率达到98.34%,在其它算法评价指标中均表现良好。可视化结果进一步验证了Efficient-TAO算法在疾病分类中的可信度。Thyroid-associated ophthalmopathy(TAO)is a common chronic blinding orbital disease with poorly characterized early pathology.To solve the problem of misdiagnosis caused by subjective diagnostic bias of physicians in the TAO diagnosis process,a model Efficient-TAO based on the compound model scaling(CMS)method was proposed.Through optimizing the training process,the dynamic ReLU activation function,stochastic weight averaging function were improved to stimulate the perfor-mance.Direct training of ocular muscle CT image data was achieved by using image pre-processing algorithms.Experimental results show that the model achieves 98.34%classification accuracy and performs well in other algorithm evaluation metrics.The visualization results further validate the credibility of the Efficient-TAO algorithm for TAO classification.

关 键 词：甲状腺相关眼病 Efficient-TAO 复合模型缩放 模型优化 医学图像分类 图像预处理 可信度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]