Mitsuteru Tsuchiya1, Takayuki Masui2, Kazuma Terauchi2, Takahiro Yamada2, Motoyuki Katyayama2, Shintaro Ichikawa3, Yoshifumi Noda4, Satoshi Goshima3. 1. Department of Radiology, Seirei Hamamatsu General Hospital, 2-12-12, Sumiyoshi, Naka-ku, Hamamatsu city, Shizuoka, 430-8558, Japan. tsuchi8@hama-med.ac.jp. 2. Department of Radiology, Seirei Hamamatsu General Hospital, 2-12-12, Sumiyoshi, Naka-ku, Hamamatsu city, Shizuoka, 430-8558, Japan. 3. Department of Radiology, Hamamatsu University School of Medicine, 1-20-1, Handayama, Higashi-ku, Hamamatsu City, Shizuoka, 431-3192, Japan. 4. Department of Radiology, Gifu University, 1-1, Yanagido, Gifu City, Gifu, 501-1194, Japan.
Abstract
OBJECTIVES: To evaluate the diagnostic performance of MRI-based radiomics model for differentiating phyllodes tumors of the breast from fibroadenomas. METHODS: This retrospective study included 88 patients (32 with phyllodes tumors and 56 with fibroadenomas) who underwent MRI. Radiomic features were extracted from T2-weighted image, pre-contrast T1-weighted image, and the first-phase and late-phase dynamic contrast-enhanced MRIs. To create stable machine learning models and balanced classes, data augmentation was performed. A least absolute shrinkage and selection operator (LASSO) regression was performed to select features and build the radiomics model. A radiological model was constructed from conventional MRI features evaluated by radiologists. A combined model was constructed using both radiomics features and radiological features. Machine learning classifications were done using support vector machine, extreme gradient boosting, and random forest. The area under the receiver operating characteristic (ROC) curve (AUC) was computed to assess the performance of each model. RESULTS: Among 1070 features, the LASSO logistic regression selected 35 features. Among three machine learning classifiers, support vector machine had the best performance. Compared to the radiological model (AUC: 0.77 ± 0.11), the radiomics model (AUC: 0.96 ± 0.04) and combined model (0.97 ± 0.03) had significantly improved AUC values (both p < 0.01) in the validation set. The combined model had a relatively higher AUC than that of the radiomics model in the validation set, but this was not significantly different (p = 0.391). CONCLUSIONS: Radiomics analysis based on MRI showed promise for discriminating phyllodes tumors from fibroadenomas. KEY POINTS: • The radiomics model and the combined model were superior to the radiological model for differentiating phyllodes tumors from fibroadenomas. • The SVM classifier performed best in the current study. • MRI-based radiomics model could help accurately differentiate phyllodes tumors from fibroadenomas.
OBJECTIVES: To evaluate the diagnostic performance of MRI-based radiomics model for differentiating phyllodes tumors of the breast from fibroadenomas. METHODS: This retrospective study included 88 patients (32 with phyllodes tumors and 56 with fibroadenomas) who underwent MRI. Radiomic features were extracted from T2-weighted image, pre-contrast T1-weighted image, and the first-phase and late-phase dynamic contrast-enhanced MRIs. To create stable machine learning models and balanced classes, data augmentation was performed. A least absolute shrinkage and selection operator (LASSO) regression was performed to select features and build the radiomics model. A radiological model was constructed from conventional MRI features evaluated by radiologists. A combined model was constructed using both radiomics features and radiological features. Machine learning classifications were done using support vector machine, extreme gradient boosting, and random forest. The area under the receiver operating characteristic (ROC) curve (AUC) was computed to assess the performance of each model. RESULTS: Among 1070 features, the LASSO logistic regression selected 35 features. Among three machine learning classifiers, support vector machine had the best performance. Compared to the radiological model (AUC: 0.77 ± 0.11), the radiomics model (AUC: 0.96 ± 0.04) and combined model (0.97 ± 0.03) had significantly improved AUC values (both p < 0.01) in the validation set. The combined model had a relatively higher AUC than that of the radiomics model in the validation set, but this was not significantly different (p = 0.391). CONCLUSIONS: Radiomics analysis based on MRI showed promise for discriminating phyllodes tumors from fibroadenomas. KEY POINTS: • The radiomics model and the combined model were superior to the radiological model for differentiating phyllodes tumors from fibroadenomas. • The SVM classifier performed best in the current study. • MRI-based radiomics model could help accurately differentiate phyllodes tumors from fibroadenomas.
Authors: Susanne Wurdinger; Aimée B Herzog; Dorothee R Fischer; Christiane Marx; Gerd Raabe; Achim Schneider; Werner A Kaiser Journal: AJR Am J Roentgenol Date: 2005-11 Impact factor: 3.959
Authors: Fergus Davnall; Connie S P Yip; Gunnar Ljungqvist; Mariyah Selmi; Francesca Ng; Bal Sanghera; Balaji Ganeshan; Kenneth A Miles; Gary J Cook; Vicky Goh Journal: Insights Imaging Date: 2012-10-24