研究者情報

タキシタ　マリコ   TAKISHITA MARIKO   瀧下茉莉子    所属   医学部医学科 乳腺・内分泌外科    職種   診療助手
言語種別	日本語
発表タイトル	Axillary lymph node metastasis prediction with FFDM and s2D mammography based on machine learning
会議名	第30回日本乳癌学会学術総会
学会区分	全国規模の学会
発表形式	ポスター掲示
講演区分	一般
発表者・共同発表者	◎原口貴史、後藤由香、古谷悠子、瀧下茉莉子、津川浩一郎、印牧義英、小林泰之
発表年月日	2022/06/30
開催地 （都市, 国名）	パシフィコ横浜
開催期間	2022/06/30～2022/07/02
概要	Background: In recent years, the use of digital breast tomosynthesis (DBT) has increased besides full-field digital mammography (FFDM), and synthetic 2D (s2D) mammography　reconstructed from DBT was developed to overcome the problem of increased radiation dose. The purpose of this study was to establish a model with FFDM and a model with s2D mammography using texture analysis to predict axillary lymph node metastasis, thereby evaluating the performance of models. Methods: Seventy-eight patients with breast cancer were enrolled in this retrospective study. All patients underwent FFDM and DBT and s2D mammography was reconstructed from DBT. Texture features were calculated based on segmented ROIs for mass lesions using free medical image processing software (3D slicer, version 4.11). The patients were randomly divided into a training set (n=58) and a testing set (n=20). Feature selection was performed using the model-based method using the train set. Extreme gradient boosting (XGB) was used as the prediction model. The performance of the models was evaluated using the area under the curve (AUC), accuracy score, sensitivity (recall rate), specificity, and the mean AUC with cross-validation (5-fold). Results: The average AUC was 0.88 of a model with FFDM, and was 0.76 of a model with s2D. Accuracy score, sensitivity (recall rate), specificity, and the mean AUC of a model with cross-validation (5-fold) were 0.85, 0.83, 0.86, 0.89 of a model with FFDM, and 0.85, 0.67, 0.93, 0.75 of a model with s2D mammography. Conclusions: This study revealed the performance of a model with FFDM and a model with s2D mammography and showed that s2D mammography might be effective in predicting axillary lymph node metastasis, although it is not as effective as FFDM.