Multivariate discriminant analysis of normal, intraepithelial neoplasia and human papillomavirus infection of the uterine cervix samples.

: The present investigation studies the role of multivariate statistical methods on quantitative histopathological features of cells in uterine cervix epithelium to discriminate between normal and abnormal uterine cervix samples. 143 histological specimens were included in the study involving normal cervix, cervical intraepithelial neoplasia (CIN) lesions and cervical human papillomavirus (HPV) infection with and without CIN (condyloma-CIN and condyloma-NCIN groups, respectively). Deep, middle and superficial regions of the cervical squamous epithelium were morphometrically analyzed. Identification of normal cervix from pathological cases was highly achieved with a specificity of 100%. The application of discriminant statistical method within pathological specimens showed an acceptable percentage of cases correctly classified; thus, an efficiency of 83.0% and 74.6% was obtained in order to discriminate within CIN and condyloma-CIN grades respectively. These percentages increased when differentiation between each grade of CIN versus condyloma-CIN were considered, using only 1-3 morphometrical parameters. Our findings indicate that the combination of nuclear and cytoplasmic quantitative features, specially size parameters, permit a high correct percentage classification of cervix samples. The discrimination process was better when few diagnostic categories were included; however, 100% specificity for normal samples was always reached.