DBN Classifier for Classification of Benign and Malignant Solitary Pulmonary Nodule
At present，image classification technology is more and more used in the qualitative diagnosis of medical images，which promotes the development of computer-aided diagnosis． For lung CT images of solitary pulmonary nodules ( SPN) classification problem，the main idea of the research is seeking to expre...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Harbin University of Science and Technology Publications
2018-06-01
|
| Series: | Journal of Harbin University of Science and Technology |
| Subjects: | |
| Online Access: | https://hlgxb.hrbust.edu.cn/#/digest?ArticleID=1958 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | At present，image classification technology is more and more used in the qualitative diagnosis of medical images，which promotes the development of computer-aided diagnosis． For lung CT images of solitary pulmonary nodules ( SPN) classification problem，the main idea of the research is seeking to express the characteristics of SPN images effectively through the classifier for benign and malignant SPN accurate discrimination． Before this team work，768 dimensional feature space is obtained by multi-resolution histograms of SPN image． Features for Support vector machine ( SVM) classifier is trained，and finally achieved the satisfactory classification results． This paper is combined with the field of deep-learning related knowledge，the deep-learning of the Deep Belief Network ( DBN) for SPN benign and malignant classification task． First of all，the CT image is segmented from the lung SPN image as the input of DBN for unsupervised training; Then，the final DBN model is obtained by fine tuning of the SPN image with benign and malignant classification． Finally，the trained DBN model is used to classify the test image set． In the part of experiment，480 cases of pulmonary nodules were selected，and 600 SPN images were extracted as the experimental data． The proposed DBN model and SVM model based on the texture features and multi-resolution histogram features were compared in the absence of consideration of the medical symbol，the DBN model for high identification accuracy of 86% ，the classification performance of a SVM classifier have significantly improved． |
|---|---|
| ISSN: | 1007-2683 |