宫颈上皮细胞图象的计算机分类

通过对约300个宫颈上皮细胞图象的分析研究，结果表明:(1)当将对象分为三类(正常、 核异质与癌)时,三级线性判决的效果明显地优于一级判决;(2)从纯数学角度选取的非视觉特 征对分类是重要的.在适当地选择训练集且限定每次判决所用特征数为五时,正确识别率可 达75%.

The Computerized Sorting for Cervical Epitheliums

About 300 cervical epithelium images were analysed ,and studied as a whole. The results have shown that 1) when these images are devided into three classes (normal, dysplastic, cancer), tertiary linear decision effect is much better than the primary level decision and 2) non-visual features that is acquired from pure-mathematical method are important for sorting. When numbers of feature to each decision is restricted to five and training sets are preperly selected, the right recognition ratio achieved is 75%.