Radiomics-based cerebrospinal fluid cell classification(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2023年第2期
- Page:
- 244-250
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- Radiomics-based cerebrospinal fluid cell classification
- Author(s):
- QIAO Lin1; WU Wenna2; LU Zhentai2
- 1. Clinical Laboratory, Guangdong Sanjiu Brain Hospital, Guangzhou 510515, China 2. School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
- Keywords:
- white cerebrospinal fluid cell segmentation shape and color features radiomics feature extraction
- PACS:
- R318；R331.142
- DOI:
- DOI:10.3969/j.issn.1005-202X.2023.02.020
- Abstract:
- Abstract: Objective To build a more efficient and accurate cerebrospinal fluid cell discrimination model based on radiomics. Methods A total of 3 331 microscopic images of cerebrospinal fluid cells were retrospectively collected, including 167 of phagocytes, 332 of monocytes, 1 081 of lymphocytes, and 1 751 of neutrophils. After segmenting the nucleus, nucleus convex hull and some cytoplasm of the nucleus convex hull in the microscopic images, 3 kinds of nuclear shape features, including roundness, convexity and firmness, 48 color features of the nucleus, nucleus convex hull and some cytoplasm of the nucleus convex hull, and 4 676 texture features of nucleus convex hull were extracted. Results A total of 4 727 radiomic features were obtained. After ANOVA and LASSO feature selection, only 519 features were retained, and both shape features and color features were retained in a high proportion (100.0%, 66.7%). After feature selection, SMOTE data enhancement and SVM classifier were used for the prediction on the test set. The results showed that the accuracy, sensitivity, specificity, precision, F1_score, and AUC were as high as 0.953, 0.948, 0.990, 0.961, 0.955, and 0.996, respectively. Conclusion The proposed feature extraction scheme and classification model for cell microscopic image is effective for cell classification, and avoid the cytoplasm segmentation. It is not necessary to segment cells, but only to segment the nucleus and the nucleus convex hull to obtain better classification results.
Last Update: 2023-03-03