Skip to main content

Table 3 Performances for the manual model

From: MRI-based radiomics to predict response in locally advanced rectal cancer: comparison of manual and automatic segmentation on external validation in a multicentre study

  Construction set: centre A + centre B
  AUC
(95% CI)
ACC %
(95% CI)
[rate]
SE %
(95% CI)
[rate]
SP %
(95% CI)
[rate]
NPV %
(95% CI)
[rate]
PPV %
(95% CI)
[rate]
MRMR+EL (AdaBoost) 1.00
(93–100)
99
(92–100)
100
(97–100)
98
(87–100)
100
(80–99)
96
(95–100)
[66/67] [39/39] [27/28] [27/27] [39/40]
Ranking+EL (Bag) 0.99
(92–100)
94
(85–98)
97
(80–98)
90
(81–100)
96
(74–96)
93
(84–100)
[63/67] [37/38] [26/29] [26/27] [37/40]
Ranking+SVM Gaussian 0.87
(79–95)
81
(69–89)
93
(76–99)
73
(67–88)
94
(79–98)
69
(58–79)
[54/67] [25/27] [29/40] [29/31] [25/36]
Ranking+LR stepwise 0.69
(67–85)
80
(69–89)
[54/67]
89
(71–98)
[24/27]
75
(59–87)
[30/40]
91
(77–97)
[30/33]
70
(57–80)
[24/34]
Ranking+SVM polynomial 0.90
(82–97)
83
(71–90)
[55/67]
85
(66–96)
[23/27]
80
(64–91)
[32/40]
89
(76–95)
[32/36]
74
(60–84)
[23/31]
  Validation set: centre C
Ranking+SVM polynomial 0.61
(52–74)
68
(48–84)
[19/28]
60
(32–83)
[9/15]
77
(47–95)
[10/13]
63
(46–76)
[10/16]
75
(60–90)
[9/12]
  1. AUC Area under the curve, ACC Accuracy, NPV Negative predictive value, PPV Positive predictive value, SE Sensitivity, SP Specificity