基于深度学习的声带疾病诊断识别方法比较研究  

Comparative Study on Diagnosis and Recognition Methods of Vocal Cord Diseases Based on Deep Learning

作　　者：邹锋 郭珊珊 樊玉琦 ZOU Feng;GUO Shanshan;FAN Yuqi(Zhejiang Pharmaceutical University,Ningbo 315100,China;Ningbo Yinzhou No.2 Hospital,Ningbo 315192,China;Hefei University of Technology,Hefei 230009,China)

机构地区：[1]浙江药科职业大学,浙江宁波315100 [2]宁波市鄞州区第二医院,浙江宁波315192 [3]合肥工业大学,安徽合肥230009

出　　处：《现代信息科技》2024年第8期111-114,122,共5页Modern Information Technology

基　　金：浙江省医药卫生科技计划项目(2022PY090);浙江省教育厅科研项目(Y202147891);2020年宁波市鄞州区农业与社会发展科技项目。

摘　　要：在医学图像诊断领域,计算机辅助诊断技术已提升了图像诊断的准确性,但针对声带疾病的喉镜图像深度学习模型仍相对稀缺,这在一定程度上限制了声带疾病识别领域的发展。文章采用经典的VGG-Net算法和一种引入注意力机制的算法来对喉镜图像进行分类。通过比较这两种算法在准确率、召回率/灵敏率和特异率方面的表现,评估它们在医学图像分类性能上的优劣。实验结果表明,引入注意力机制的SA、SE-Net、CBAM和ECA-Net算法在性能上明显优于VGG-Net算法。结合深度学习和注意力机制可显著提升声带疾病喉镜图像诊断的准确性和效率,这对未来医疗行业的健康发展有着极其重要的意义。In the field of medical image diagnosis,computer-aided diagnostic technology has improved the accuracy of image diagnosis,but laryngoscope image Deep Learning models for vocal cord disease are still relatively scarce,which to some extent limits the development of the field of vocal cord disease recognition.This paper uses the classic VGG-Net algorithm and an algorithm that introduces Attention Mechanism to classify laryngoscope images.Evaluate the performance of these two algorithms in medical image classification by comparing their accuracy,recall/sensitivity,and specificity.The experimental results show that the SA,SE-Net,CBAM,and ECA-Net algorithms that introduce Attention Mechanisms have significantly better performance than the VGG-Net algorithm.The combination of Deep Learning and Attention Mechanisms can significantly improve the accuracy and efficiency of laryngoscopy image diagnosis for vocal cord disease,which is of great significance for the healthy development of the future medical industry.

关 键 词：医学图像诊断 声带疾病 喉镜图像 VGG-Net算法 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]