Affiliations: [a] Department of Convergence Software, Hallym University, Chuncheon, Ganwon-do, Korea | [b] Bio-IT Research Center Hallym University, Chuncheon, Ganwon-do, Korea
Correspondence:
[*]
Corresponding authors: Yu-Seop Kim, Department of Convergence Software, Hallym University; Bio-IT Research Center Hallym University, Chuncheon, Ganwon-do, 2002-702, Korea. E-mail:[email protected]
Abstract: We try to find out the best weight values of CA125 and
HE4 in a discriminant formula classifying ovarian cancer patients from
benign patients. We utilize a logistic regression analysis for the early
screening system of the ovarian cancer for Korean patients. We compare our
system with ROMA (Risk of Ovarian Malignancy Algorithm) of Abbot corp. In
view of AUC (Area under the ROC curve), sensitivity with 95% of
specificity and accuracy are considered. We performed experiments based on
the logistic regression analysis separated by the case of pre- and post-
menopausal stages and by the stages of progression of cancer. In our
experiments, we can increase about 15.6% points of sensitivity with
95% of specificity, compared to that of ROMA. In premenopausal cases,
ROMA shows 93.32% of AUC value and our system shows 97.48% of AUC,
4.1% points higher than ROMA. AUC of the ROMA for premenopausal women was
93.32%, whereas the AUC of our system was 97.48%. Furthermore, the AUC
of the ROMA for early-staged ovarian cancer was 91.35%, whereas the AUC
of our system was 97.22%, showing that the diagnostic performance of our
system was superior over that of the ROMA in Korean patient cases.
Keywords: Risk of ovarian malignancy algorithm, CA125, HE4, tumor marker, ovarian cancer, sensitivity, specificity, epithelial ovarian cancer
