基于人工神经网络模型的肺癌CT图像分割算法

An Algorithm of Lung Cancer CT Image Segmentation Based on Artificial Neural Network Model

目的提出一种基于人工神经网络模型的肺癌CT图像分割算法。方法第一步采用维纳滤波器和模糊增强抑制图像噪声和提升图像对比度,第二步提取图像的纹理特征和分形特征,第三步根据网络最佳参数训练和测试人工神经网络模型,第四步提取CT图像中肺癌病灶区域。512个样本和80例图像被用来训练和测试模型。结果肺癌CT图像包含13个癌症显著区域特征(3个纹理特征和10分形特征)。训练和测试数据所得最佳分类函数为列文伯格-马夸尔特反向传播,学习速率R为0.3,动量为0.9,隐藏神经元数量为20。训练阶段灵敏度、特异度和准确度可达98.4%,100%和98.6%,同时测试阶段对应指标分别可达90.9%,100%和95.1%。结论基于人工神经网络模型的图像分割算法能高效、准确定的提取CT肺癌病灶,可作为影像医师诊断肺癌的有效工具。

Objective To develop a lung cancer CT image segmentation algorithm based on artificial neural network model. Methods The first step was to use Wiener filter and fuzzy enhancement to suppress image noise and enhance image contrast. The second step was to extract texture and fractal features of the image. The third step was to train and test the artificial neural network model according to the best parameters of the network. The fourth step was to extract the lung cancer lesion area in CT image. A total of 512 samples and 80 images were used to train and test the model. Results CT images of lung cancer contained 13 significant regional features(including 3 texture features and 10 fractal features). The best classification function obtained from training and testing data was Levenberg-Marquart back propagation. The learning rate R was 0.3, the momentum was 0.9, and the number of hidden neurons was 20. The training sensitivity, specificity and accuracy were 98.4%, 100% and 98.6%, respectively. The corresponding indicators in the test stage could reach 90.9%, 100% and 95.1%, respectively. Conclusion Image segmentation algorithm based on artificial neural network model can extract lung cancer lesions efficiently and accurately, and can be used as an effective tool for diagnosing lung cancer.