Int J Performability Eng ›› 2019, Vol. 15 ›› Issue (1): 326-336.doi: 10.23940/ijpe.19.01.p33.326336

Previous Articles     Next Articles

Fully Convolutional-based Dense Network forLungNodule Image Retrieval Algorithm

Pinle Qinab, Qi Liab, Jianchao Zengab*(), Haiyan Liuc, and Yuhao Cuia   

  1. a School of Data Science and Technology,North University of China,Taiyuan,030051,China
    b ShanxiProvincialKeyLaboratoryforBiomedicalImagingandBigData,Taiyuan,030051, China
    c First Hospital of Shanxi Medical University,Taiyuan,030051, China
  • Revised on ; Accepted on
  • Contact: Zeng Jianchao E-mail:zjc@nuc.edu.cn
  • About author:Pinle Qin received the PhD degree in computer application technology from Dalian University of Technology (DLUT), Dalian, Liaoning, P.R. China, in 2008. He is currently an associate professor with the School of Data Science and Technology, North University of China (NUC). His current research interests include computer vision, medical image processing and deep learning.|Qi Li is pursing Master degree from NUC, and his areas of interest are digital image processing, medical image processing and computer vision.|Jianchao Zeng received the PhD degree from Xi’an Jiaotong University, Xi’an, Shanxi, P.R. China, in 1990. He is currently the vice president of North University of China (NUC). His current research interests include computer vision, medical image processing and deep learning.|Haiyan Liu Department of Nuclear Medicine, First Hospital of Shanxi Medical University, Taiyuan, Shanxi 030001, China; Molecular Imaging Precision Medical Collaborative InnovationCenter, Shanxi Medical University, Taiyuan, Shanxi 030001, China.|Yuhao Cui born in 1996, B.S. candidate. His research interests include deep learning, digital image processing.

Abstract:

At present, there are many problems in the existing content-based medical image retrieval (CBMIR) algorithms. The most important problem is the lack of feature extraction, resulting in the imperfect expression of semantic information and the lack of data-based learning ability. Meanwhile, the characteristic dimension is high, which affects the performance of image retrieval. In order to solve these problems, this paper presents a fully convolutional dense network (FCDN) algorithm, which solves the gap between the extracted low-level features and high-level semantic features. In order to improve the accuracy and efficiency of retrieval, the concept of Joint distance is proposed in this paper. Since the image information of lung nodules extracted from different layers of the network is different, the minimum Joint distance is selected by comparing the minimum Hamming distances of the layers 4, 17 and 25 of the similar images retrieved. Compared with other methods, the average accuracy of the lung nodule image retrieval can reach 91.17% under the 64-bit hash code length, the average time for retrieving a lung slice is 4.8×10 -5s,The search results not only express the rich semantic features of the image, but also improve the retrieval efficiency. And the retrieval performance is better than other network structures to help doctors assist in diagnosis.

Key words: lung nodules, medical image retrieval, fully convolutional neural networks, dense blocks, hash function