Abstract
This study aimed to explore a discriminant method for cervical squamous epithelial cells based on computer image analysis to establish a foundation for artificial intelligence diagnosis of cervical cancer. A total of 1682 cells were captured from 53 Papanicolaou smears, and computer image analysis was used to test the chromatics and geometric structural parameters of the above cells. Stepwise discriminant analysis was used to establish the discriminant function, and the discriminant effects of the function were evaluated. The chromaticity and geometric features of the cell images had significantly different among low-grade cells (LGC), atypical squamous cells of undetermined significance, and high-grade cells (HGC), as well as between subclassifications within LGC and HGC (P < .05). Stepwise discriminant analysis was used to create functions for classifying cells into the categories LGC, atypical squamous cells of undetermined significance, and HGC, as well as subclassifications within LGC and HGC. Functions combined with chromatics and geometry features have a good discriminant effect on cervical squamous cells. The discriminant coincidence rates indicate that this method can be an appropriate reference approach for the classification and diagnosis of cervical squamous epithelial cells, and its potential applications are presented in a tentative study on automated image analysis systems for cytological fields.