基于改进VGG16的乳腺肿块识别
Breast mass recognition based on improved VGG16
投稿时间:2021-01-05  修订日期:2021-01-05
DOI:
中文关键词: 乳腺肿块  识别  VGG16  深度学习
英文关键词: breast mass  recognition  VGG16  deep learning
基金项目:湖北省自然科学基金资助项目(2020CFB541); 中央高校基本科研业务费专项资金项目(CZY19011)
作者单位E-mail
盘安思 中南民族大学计算机科学学院 690055898@qq.com 
程时宇 中南民族大学计算机科学学院  
佘逸飞 中南民族大学计算机科学学院  
徐胜舟 中南民族大学计算机科学学院 xushengzhou@scuec.edu.cn 
摘要点击次数: 23
全文下载次数: 0
中文摘要:
      乳腺肿块的自动识别在乳腺癌的早期诊断中具有重要的临床应用价值.本文针对从乳腺X线摄片中提取的感兴趣区域的特点,对VGG16进行改进,减少模型中卷积层和卷积核的个数,提出一种精简的卷积神经网络模型,用于感兴趣区域中肿块的识别.同时,为避免网络模型受小样本量限制出现过拟合现象,通过旋转与翻转操作对感兴趣区域进行数据增强。实验中采用来自DDSM数据集的703个感兴趣区域作为训练集,另外176个感兴趣区域作为测试集.采用准确度、精确度、敏感度以及F1_score等指标对网络模型的性能进行评估.实验结果表明,改进的VGG16的上述评价指标分别达到了90.34%、89.87%、88.75%和0.89,明显优于其它已有的卷积神经网络模型,同时其计算效率也明显高于原始VGG16模型.
英文摘要:
      The automatic recognition of breast masses has important clinical application value in the early diagnosis of breast cancer. In this paper, based on the characteristics of the region of interest extracted from mammography, VGG16 is improved by reducing the number of convolutional layers and convolutional kernels, and a simplified convolutional neural network is proposed for the recognition of masses in the region of interest. At the same time, to avoid the over-fitting of the network caused by the limitation of small sample size, the data augmentation is carried out by rotation and flip operation. In the experiment, 703 regions of interest from the DDSM dataset are used as the training set, and the other 176 regions of interests are used as the test set. Using the accuracy, precision, sensitivity and F1_score to evaluate the performance of the model, the experimental results show that the indexes above of the improved VGG16 reach 90.34%, 89.87%, 88.75 and 0.89 respectively, which are significantly better than other existing convolutional neural networks, and its computational efficiency is also significantly higher than the original VGG16 model.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭