Classification of liver cirrhosis using global and local features(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2023年第11期
- Page:
- 1362-1369
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Classification of liver cirrhosis using global and local features
- Author(s):
- ZHU Jiaqi1; LIU Xiang1; SONG Jialin2
- 1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China 2. Department of Ultrasound Diagnosis and Treatment, Shanghai Changzheng Hospital, Naval Medical University, Shanghai 200003, China
- Keywords:
- Keywords: liver cirrhosis computer-aided diagnosis ultrasound image processing feature analysis
- PACS:
- R318；R657.31
- DOI:
- DOI:10.3969/j.issn.1005-202X.2023.11.008
- Abstract:
- Abstract: Objective To propose a classification method for liver cirrhosis diagnosis using both global and local features of liver tissues, thereby overcoming the current problem that the diagnosis of liver cirrhosis mainly relies on manual detection. Methods The two-dimensional high-frequency ultrasound images were collected from 47 patients with hepatitis B cirrhosis (divided into mild, moderate and severe cirrhosis groups according to Child-Pugh score) and 20 healthy volunteers (normal control group). The differential filter was used to obtain the code distance of the image, and then detect the corner point, cord, ascites and other features for quantitative analysis. A two-stage classification model was established. The support vector machine (SVM) of radial basis function was used in the first stage of classification to group mild and moderate cirrhosis into one category, while normal controls and severe cirrhosis into the other category for avoiding mutual interference between the two categories. In the second stage, after parameters tuning with Bayesian, random forest (RF) algorithm was used to adjust the importance weights of different features for further enhancing the classification accuracy of liver cirrhosis. Results All of the selected features were statistically significant (P<0.05), and the SVM-RF model had an accuracy of 93.11%, 88.19%, 91.93% and 96.86% for diagnosing normal liver, mildly cirrhosis liver, moderate cirrhosis liver and severe cirrhosis liver. Conclusion The proposed SVM-RF method can effectively extract the global and local features that conform to the doctors perspective, and assist in cirrhosis diagnosis.
Last Update: 2023-11-24