Skip to main content

Table 2 Baseline characteristics of the included studies

From: A meta-analysis of the diagnostic performance of machine learning-based MRI in the prediction of axillary lymph node metastasis in breast cancer patients

Author (year of publication) Algorithm Sequences Segmentation Dataset Sensitivity Specificity
Hongna Tan (2020) SVM T2-FS 3D Training set 84.38% 72.25%
     Validation set 65.22% 81.08%
Thomas Ren (2020) CNN T1CE 2D Validation set 92.10 ± 2.90% 79.30 ± 5.10%
Xiao Zhang (2019) RF DWI 3D Validation set 83.30% 74.20%
   T2-FS    77.80% 87.10%
Jia Liu (2019) SVM T1CE 3D Training set 75.00% 76.00%
     Validation set 71.00% 100.00%
  Xgboost    Training set 89.00% 76.00%
     Validation set 86.00% 83.00%
  LR    Training set 71.00% 71.00%
     Validation set 71.00% 83.00%
Karl D Spuhler (2019) CNN T1CE 3D Testing set 72.20% 88.90%
Lu Han (2019) SVM T1CE 2D Training set 89.00% 57.00%
     Validation set 78.00% 72.00%
Jiaxiu Luo (2018) SVM DWI 3D Testing set 86.70% 83.30%
Chunling Liu (2019) LR T1CE 3D Training set 76.10% 66.70%
     Validation set 81.90% 77.80%
Yuhao Dong (2018) LR T2-FS 3D Training set 66.30 ± 0.30% 81.60 ± 0.20%
     Validation set 60.00 ± 0.60% 74.70 ± 0.40%
   DWI   Training set 74.00 ± 0.20% 80.80 ± 0.20%
     Validation set 69.50 ± 0.50% 75.70 ± 0.40%
   T2-FS and DWI   Training set 66.30 ± 0.30% 81.60 ± 0.20%
     Validation set 70.00 ± 0.80% 74.70 ± 0.50%
Meijie Liu (2020) LR T1CE 2D Validation set 64.00% 79.00%
Xiaoyu Cui (2019) SVM T1CE 3D Validation set 94.90% 77.96%
  KNN     89.39% 87.18%
  LDA     80.31% 67.78%
Demircioglu (2020) LR T1CE and T2-FS 3D Validation set 71.00% 74.00%
Arefan (2020) LDA T1CE 2D Testing set 60.00% 87.00%
  RF     60.00% 86.00%
  NB     81.00% 67.00%
  KNN     74.00% 54.00%
  SVM     70.00% 71.00%
  LDA   3D   63.00% 92.00%
  RF     64.00% 90.00%
  NB     85.00% 62.00%
  KNN     67.00% 58.00%
  SVM     81.00% 50.00%
  LDA   2D Validation set 82.00% 76.00%
  RF     89.00% 68.00%
  NB     73.00% 82.00%
  KNN     79.00% 75.00%
  SVM     65.00% 88.00%
  LDA   3D   82.00% 78.00%
  RF     89.00% 70.00%
  NB     69.00% 76.00%
  KNN     75.00% 75.00%
  SVM     73.00% 79.00%
Fusco (2018) LDA T1CE 3D Validation set 88.50% 77.80%
  1. CNN, convolutional neural networks; SVM, support vector machine; LR, linear regression; KNN, k-nearest neighbor; LDA, linear discriminant analysis; NB, naive Bayes; RF, random forest; T2-FS, fat-suppressed T2; T1CE, contrast-enhanced T1. The algorithm pooled was chosen with the highest performance