基于视觉词袋模型的亚损伤红细胞识别

根据亚损伤红细胞的形态学变化，提出一种自动识别亚损伤红细胞的方法。采用体外循环过程中的血细胞图像，包括2 763张亚损伤红细胞图像和2 507张正常红细胞，利用视觉词袋作为红细胞特征提取方法，分别选用多项核、高斯核、Sigmoid核函数的支持向量机模型。采用5折交叉验证方法验证方法的性能，并选取精确度、召回率、F1评分作为评价指标。结果表明3种不同内核模型的识别准确率分别为91.05%±0.82%、94.16%±0.50%、85.60%±0.94%。本研究提出的方法有效区别了亚损伤红细胞，为亚致死性损伤检测提供自动化方案。

Identification of sublethally damaged red blood cells based on bag-of-visual words model

Abstract: A method for the automatic identification of sublethally damaged red blood cells (RBC) according to their morphological changes is proposed. The images of blood cells during cardiopulmonary bypass are analyzed in the study, including 2 763 images of sublethally damaged RBC and 2 507 images of normal RBC. The bag-of-visual words is used as the feature extraction method of RBC, and the support vector machine models of polynomial kernel, Gaussian kernel and Sigmoid kernel function are selected separately. A 5-fold cross validation method is adopted to verify the performance of the method, taking precision rate, recall rate and F1 score as evaluation indexes. The results show that the identification accuracies under 3 different kernel models are 91.05%±0.82%, 94.16%±0.50% and 85.60%±0.94%, respectively. The proposed method can effectively distinguish the sublethally damaged RBC and provide a scheme for the automatic identification of sublethally damaged RBC.