基于深度卷积神经网络的肺结节检测与识别

应用卷积神经网络将肺结节从含有背景、噪声的胸腔区域里检测并识别出来。首先,对图像进行预处理,获得肺实质图像。然后,应用Faster R-CNN多特征融合算法检测肺结节候选区域,再利用多角度特征融合方法滤除假阳性结节。接着,通过数据增强法、残差学习法、优化初始参数等对卷积神经网络的性能进行优化。最后,应用迁移学习方法对数据集进行训练,得出最终的检测结果。抽取LIDC数据集中含有肺结节图像数据,检测并识别肺结节的准确率达到98. 1%。实验结果表明,该算法优于其他3类算法,实现了肺结节的精确检测和识别,在保证检测和识别出正确结节的前提下,降低了过拟合率及训练时间,提高了算法效率,研究成果为早期肺癌的诊断提供参考依据。

Lung Nodules Detection and Recognition based on Deep Convolutional Neural Networks

Objective (Convolutional neural network, CNN) was used to detect and recognize lung nodules from the thoracic region of containing background and noise. Methods Firstly, Image preprocessing and Obtain images of lung parenchyma. Secondly, Faster R-CNN multi-feature fusion algorithm was applied to detect candidate regions of lung nodules, False-positive nodules were filtered by multi-angle feature fusion method. Then the performance of convolutional neural network was optimized by data enhancement method, residual learning method, optimization of initial parameters. Finally, the transfer learning method was applied to train the dataset, and test results were expected. Results The containing lung nodules image data were extracted from LIDC dataset, Detection and recognition accuracy of lung nodules reached 98.1%. The experimental results showed that the performance of this algorithm was better than other three algorithms algorithm and can accurately detect and recognize lung nodules. Conclusion Detection and recognize accurately of lung nodules on the premise that the overfitting rate and training time were reduced and the algorithm efficiency was improved, The results could provide reference for the diagnosis of early lung cancer.