基于医学征象和卷积神经网络的肺结节CT图像哈希检索

摘要

针对肺结节图像检索中存在的两个问题:手工设计的特征对肺结节的表达能力不强,生成的哈希码检索效果不佳.文中提出一种基于医学征象和卷积神经网络的肺结节CT图像哈希检索方法.首先,依据肺结节的9种征象取值,构造训练集准确的哈希码;其次,利用卷积神经网络和主成分分析法提取肺结节的重要语义特征,并结合训练集准确的哈希码反向求解哈希函数;最后,提出一种基于自适应比特位的检索方法,实现待查询肺结节图像的快速检出.通过对数据集进行实验和分析,证实了本文方法在肺结节图像检索过程中取得了较高的准确率和检索精度.%Existing pulmonary nodule retrieval methods have two problems; it is difficult to express the characteristics of pulmonary nodules using hand-crafted features and the generated hashing codes have poor retrieval performance. To address these issues, this paper proposes a retrieval method for pulmonary nodules in CT images based on medical signs and convolutional neural networks. We first constructed accurate hashing codes using an accurate training set based on the values of the nine signs of pulmonary nodules. We then extracted the important semantic features of pulmonary nod-ules using convolutional neural networks and principal components analysis. In addition, we inversely solved the hash-ing functions by combining the hashing codes with the accurate training set. Finally, we developed a retrieval method, based on adaptive bits, to achieve fast searching for pulmonary nodule images. Extensive experiments and evaluations on data sets show that the method has high accuracy and retrieval precision in the process of pulmonary nodule image retrieval.