基于特征融合的新冠病毒感染肺部CT图像分类
Classification of CT Images of COVID-19 Infected Lungs Based on Feature Fusion
朱镕 1潘伟 1史润发 1胡国华 1连顺 2梅腱1
作者信息
- 1. 合肥学院先进制造工程学院,合肥 230601
- 2. 科大讯飞股份有限公司,合肥 230088
- 折叠
摘要
针对难以收集大量标注的高质量医学图像造成医学图像分类困难的问题,提出了一种基于迁移学习和特征融合的新冠病毒感染肺部CT图像分类方法,旨在提高图像分类的准确率与分类速度.通过预处理和数据增强技术滤除无用特征,采用注意力模块更好地挖掘深层次的特征信息,同时运用一种二分类的焦点损失函数来解决数据集分布不平衡的问题.实验结果表明,所用方法的图像分类准确率可达97.79％,相比较单个模型,准确率分别上升了 2.61％和1.81％,有效地提高了新冠病毒感染肺部CT图像分类准确率;同时该分类模型具有较好的泛化性能,为提高医学图像分类准确率提供有效支持.
Abstract
This paper proposes a CT image classification method for COVID-19-infected lungs based on transfer learning and feature fusion,aiming to improve the accuracy and speed of image classification in the face of the difficulty in collecting a large number of well-annotated medical images.By using preprocessing and data enhancement techniques to filter out useless features,the method employs an attention module to better extract deep-level feature information,and uses a binary Focal Loss function to address the problem of imbalanced dataset distribution.Experimental results show that the proposed method achieves an image classification accuracy of 97.79％,representing an improvement in accuracy of 2.61％ and 1.81％compared to single models,effectively improving the accuracy of COVID-19-infected lung CT image classification.The classification model also exhibits good generalization performance,providing effective support for improving the accuracy of medical image classification.
关键词
深度学习/迁移学习/特征融合/图像分类/肺部CT识别Key words
deep learning/transfer learning/feature fusion/image classification/lung CT rec-ognition引用本文复制引用
基金项目
安徽省高等学校自然科学研究项目(KJ2019A0838)
高等学校省级质量工程项目(2022)(2022sx128)
出版年
2024
NETL
NSTL
万方数据