基于CT影像数据和改进降噪自编码器的肺结节生长预测模型

Prediction of lung nodule growth based on CT image data and improved noise reduction self encoder

为提高患者肺结节疾病诊断的准确率,提出一种基于改进降噪自编码器(denoising autoencoder,DAE)的三维肺结节CT图像预测方法。其中,采用面绘制法对肺结节CT图像进行三维体素重建,以获取三维肺结节图像,采用热核特征表示方法表征三维肺结节低层特征,最后采用改进的DAE网络对肺结节进行预测,以实现肺结节CT影像的精准预测。仿真结果表明,相较于标准的DAE模型和深度信念网络(deep belief network,DBN)、金字塔匹配(intrinsic pyramid matching,ISPM)、降调距离矩阵方法(reduced Bi-harmonic distance matrix,RBiHDM),本研究提出的方法对二维数据和三维数据的识别准确率分别达到81.2%和96.2%,具有更高的准确性。由此得出,改进降噪自编码器可识别肺结节影像,更好的辅助医师诊断。

In order to improve the accuracy of lung nodule prediction,a method based on improved denoising auto encoder(DAE)model for lung nodule CT image prediction was proposed.Methods Taking the CT images of pulmonary nodules as the research object,the 3D voxel reconstruction of the CT images of pulmonary nodules was carried out by using the surface rendering method to obtain the 3D images of pulmonary nodules,and the lower layer characteristics of the 3D pulmonary nodules were characterized by using the thermonuclear feature representation method.Finally,the improved DAE model was used to predict the CT images of pulmonary nodules.The simulation results show that the proposed method can realize the prediction of pulmonary nodule disease and assist doctors in diagnosing pulmonary nodules.Compared with the standard DAE model,deep belief network(DBN),pyramid matching(ISPM),and reduced Bi harmonic distance matrix(RBiHDM),the proposed method has stronger accuracy,with the recognition accuracy of 81.2%and 96.2%for 2D data and 3D data,respectively,which has strong effectiveness and superiority.