加权Fisher卷积神经网络乳腺癌检测模型

A Model Based on Weighted Fisher LetNet-5 Criterion Convolutional Neural Network for Breast Cancer Detection

乳腺癌是女性发病率最高的癌症,已经严重威胁到女性生命健康。随着医院影像数据量的爆炸式增长和计算机图像分类技术的不断涌现,基于人工智能技术的医学图像分类也逐渐应用于大型医院的科研、临床应用中。将这种手段运用于乳腺癌影像检测中,对其辅助诊断和病理研究有着重要的意义。该文提出一个新的代价函数加权Fisher准则并将其应用于乳腺癌检测的卷积神经网络模型中,其目的在于保证图像输出值和样本标签之间的残差最小的同时,使得同类样本距离越近越好,异类样本距离越远越好,从而加强模型对乳腺癌的分类能力。在公开数据集和医院提取的真实数据上的实验表明,加权Fisher准则能有效提升卷积神经网络的收敛时间和识别率,同时基于改进的LetNet-5相较改进的AlexNet有更优的效果。将加权Fisher的LetNet-5卷积神经网络模型用于乳腺癌辅助诊断,具备一定的临床价值和应用前景。

Breast cancer is the cancer with the highest incidence in women and has seriously threatened women’s life and health.With the explosive growth of health related image data and the continuous emergence of computer image classification technology,medical image classification based on artificial intelligence technology has been gradually applied in the scientific research and clinical applications in hospitals.The application of artificial intelligence in breast cancer imaging detection shows great value in auxiliary diagnosis and pathological research.We present a new model for breast cancer detection.In this model,weighted Fisher criterion is used in the LetNet-5 convolutional neural network,which aims to ensure the minimum residual between the output value of image and the sample label,and make the distance between similar samples as close as possible,and the distance between different samples as far as possible,so as to strengthen the classification ability of the model for breast cancer.Experiments on public data sets and real data extracted from hospitals show that the weighted Fisher criterion can effectively improve the convergence time and recognition rate of convolutional neural networks.At the same time,the improved LetNet-5 is better than the improved AlexNet.The weighted Fisher’s LetNet-5 convolutional neural network model is used for breast cancer auxiliary diagnosis,which has certain clinical value and application prospects.