Table 3 The performance of the different classifiers on three independent validation cohorts.

From: Noninvasive imaging biomarker reveals invisible microscopic variation in acute ischaemic stroke (≤ 24 h): a multicentre retrospective study

Cohorts

Model

AUROC

AUPRC

ACC

SEN

SPE

PPV

NPV

F1-score

Cohort 6

RBF-SVM

0.846 [0.808 to 0.884]

0.854 [0.798 to 0.896]

0.789 (322/408) [0.746 to 0.828]

0.814 (166/204) [0.760 to 0.867]

0.765 (156/204) [0.706 to 0.823]

0.776 (166/214) [0.720 to 0.832]

0.804 (156/194) [0.748 to 0.860]

0.794

Linear-SVM

0.838 [0.798 to 0.878]

0.856 [0.801 to 0.898]

0.789 (322/408) [0.746 to 0.828]

0.779 (159/204) [0.723 to 0.836]

0.799 (163/204) [0.744 to 0.854]

0.795 (159/200) [0.739 to 0.851]

0.784 (163/208) [0.728 to 0.840]

0.787

LR

0.831 [0.790 to 0.872]

0.840 [0.783 to 0.884]

0.782 (319/408) [0.739 to 0.821]

0.696 (142/204) [0.633 to 0.759]

0.868 (177/204) [0.821 to 0.914]

0.840 (142/169) [0.785 to 0.895]

0.741 (177/239) [0.685 to 0.796]

0.761

RF

0.832 [0.791 to 0.872]

0.846 [0.789 to 0.889]

0.784 (320/408) [0.741 to 0.823]

0.716 (146/204) [0.654 to 0.778]

0.853 (174/204) [0.804 to 0.902]

0.830 (146/176) [0.774 to 0.885]

0.750 (174/232) [0.694 to 0.806]

0.768

MLP

0.838 [0.799 to 0.878]

0.827 [0.769 to 0.873]

0.784 (320/408) [0.741 to 0.823]

0.750 (153/204) [0.691 to 0.809]

0.819 (167/204) [0.766 to 0.872]

0.805 (153/190) [0.749 to 0.862]

0.766 (167/218) [0.710 to 0.822]

0.777

Cohort 7

RBF-SVM

0.805 [0.722 to 0.889]

0.804 [0.675 to 0.890]

0.764 (81/106) [0.672 to 0.841]

0.755 (40/53) [0.639 to 0.871]

0.774 (41/53) [0.661 to 0.886]

0.769 (40/52) [0.655 to 0.884]

0.759 (41/54) [0.645 to 0.873]

0.762

Linear-SVM

0.775 [0.687 to 0.864]

0.767 [0.636 to 0.862]

0.736 (78/106) [0.641 to 0.817]

0.717 (38/53) [0.596 to 0.838]

0.755 (40/53) [0.639 to 0.871]

0.745 (38/51) [0.625 to 0.865]

0.727 (40/55) [0.610 to 0.845]

0.731

LR

0.775 [0.687 to 0.864]

0.766 [0.634 to 0.861]

0.736 (78/106) [0.641 to 0.817]

0.717 (38/53) [0.596 to 0.838]

0.755 (40/53) [0.639 to 0.871]

0.745 (38/51) [0.625 to 0.865]

0.727 (40/55) [0.610 to 0.845]

0.731

RF

0.818 [0.737 to 0.898]

0.802 [0.674 to 0.889]

0.783 (83/106) [0.692 to 0.857]

0.868 (46/53) [0.777 to 0.959]

0.698 (37/53) [0.575 to 0.822]

0.742 (46/62) [0.633 to 0.851]

0.841 (37/44) [0.733 to 0.949]

0.800

MLP

0.806 [0.722 to 0.890]

0.819 [0.692 to 0.901]

0.783 (83/106) [0.692 to 0.857]

0.830 (44/53) [0.729 to 0.931]

0.736 (39/53) [0.617 to 0.855]

0.759 (44/58) [0.648 to 0.869]

0.812 (39/48) [0.702 to 0.923]

0.793

Cohort 8

RBF-SVM

0.754 [0.714 to 0.795]

0.733 [0.677 to 0.782]

0.705 (385/546) [0.665 to 0.743]

0.769 (210/273) [0.719 to 0.819]

0.641 (175/273) [0.584 to 0.698]

0.682 (210/308) [0.630 to 0.734]

0.735 (175/238) [0.679 to 0.791]

0.723

Linear-SVM

0.763 [0.723 to 0.803]

0.732 [0.676 to 0.781]

0.725 (396/546) [0.686 to 0.762]

0.696 (190/273) [0.641 to 0.751]

0.755 (206/273) [0.704 to 0.806]

0.739 (190/257) [0.686 to 0.793]

0.713 (206/289) [0.661 to 0.765]

0.717

LR

0.754 [0.713 to 0.795]

0.730 [0.674 to 0.779]

0.714 (390/546) [0.674 to 0.752]

0.736 (201/273) [0.684 to 0.789]

0.692 (189/273) [0.638 to 0.747]

0.705 (201/285) [0.652 to 0.758]

0.724 (189/261) [0.670 to 0.778]

0.720

RF

0.774 [0.735 to 0.813]

0.770 [0.716 to 0.816]

0.720 (393/546) [0.680 to 0.757]

0.751 (205/273) [0.700 to 0.802]

0.689 (188/273) [0.634 to 0.744]

0.707 (205/290) [0.655 to 0.759]

0.734 (188/256) [0.680 to 0.788]

0.728

MLP

0.731 [0.689 to 0.773]

0.693 [0.636 to 0.745]

0.689 (376/546) [0.648 to 0.727]

0.777 (212/273) [0.727 to 0.826]

0.601 (164/273) [0.543 to 0.659]

0.660 (212/321) [0.609 to 0.712]

0.729 (164/255) [0.671 to 0.787]

0.714

  1. Abbreviation: RF, random forest; SVM, support vector machine; RBF, radial basis function; LR, logistic regression; MLP, multilayer perceptron; AUROC, the area under the receiver operating characteristic curve; ACC, accuracy; SEN, sensitivity; SPE, specificity; PPV, positive predictive value; NPV, negative predictive value; AUPRC, the area under the precision recall curve.
Back to article page