Stepwise Discriminant Analysis of Liquid-Based Cytology Based on Nucleus Morphological and Colorimetric Parameters

OBJECTIVE: To explore the application value of the discriminant function in liquid-based cervical cytology diagnosis. STUDY DESIGN: Liquid-based cytology images from 34 cases of normal subjects and 224 patients diagnosed with cervical intraepithelial neoplasia were collected, and morphological and colorimetric parameters of each nucleus were measured by image analysis software. Modeling stepwise discriminant functions, we calculated the corresponding cross-validation accuracy based on the measured features of normal cells, lesion cells, and 4 different types of lesion cells, respectively. RESULTS: Each morphological and colorimetric feature between normal cells and lesions cells were statistically significant (p <0.01), and cross-validation accuracy was 88.3%. Similarly, those 2 features among 4 different types of lesion cells were also statistically significant (p <0.01), cross-validation accuracy between category 1 (atypical squamous cells of undetermined significance LASC-US] and low-grade squamous intraepithelial lesions [LSIL]) and category 2 (atypical squamous cells cannot exclude HSIL [ASC-H] and high-grade squamous intraepithelial lesions (HSIL]) was 86.1%. CONCLUSION: Using discriminant function, a diagnostic accuracy rate of at least 86.1% in the classification of ASC-US and LSIL, ASC-H and HSIL, suggested that discriminant analysis had important value in the diagnosis of liquid-based cytology.