人工知能援用による卵巣腫瘍術前診断支援システムの開発

抄録

It is difficult to diagnose benign, borderline, or malignant nature of ovarian tumors preoperatively. Problems often arise because physicians determine the extent of resection based on presumed diagnosis. There is an urgent need to improve the accuracy of preoperative diagnosis. In this study, we developed a three-level classification system for serous ovarian tumors using blood test data and machine learning. The effectiveness of combining data sets and training models was examined by learning and estimating for four different data sets and a control group using three different models: random forest, logistic regression analysis, and linear SVC. The results showed that only LR-D4 outperformed the control group in all categories, but other combinations improved in some category. This suggests that the content and combination of the data set are effective in improving precision and specificity of the estimation.