Table 2 Performance metric values obtained with six pre-trained DL models in the classification of FCD-II using two different MRI modes.

From: Focal cortical dysplasia (type II) detection with multi-modal MRI and a deep-learning framework

DL Models

PMV

MRI modes

FLAIR

T1w

Iacs

ax

cr

sg

Iacs

ax

cr

sg

VGG16

Ac

0.793 ± 0.240

[0.625, 0.960]

0.845 ± 0.106

[0.770, 0.917]

0.835 ± 0.089

[0.772, 0.897]

0.858 ± 0.097

[0.790, 0.925]

0.884 ± 0.188

[0.754, 1.000]

0.902 ± 0.153

[0.796, 1.000]

0.862 ± 0.105

[0.789, 0.935]

0.847 ± 0.188

[0.717, 0.978]

Pe

0.846 ± 0.142

[0.747, 0.945]

0.850 ± 0.109

[0.774, 0.926]

0.849 ± 0.070

[0.801, 0.898]

0.862 ± 0.094

[0.797, 0.927]

0.912 ± 0.116

[0.831, 0.993]

0.907 ± 0.141

[0.809, 1.000]

0.868 ± 0.108

[0.794, 0.943]

0.881 ± 0.125

[0.794, 0.969]

F1

0.774 ± 0.293

[0.571, 0.977]

0.843 ± 0.105

[0.769, 0.916]

0.831 ± 0.095

[0.765, 0.898]

0.857 ± 0.098

[0.789, 0.925]

0.879 ± 0.205

[0.727, 1.000]

0.901 ± 0.156

[0.793, 1.000]

0.861 ± 0.105

[0.788, 0.934]

0.841 ± 0.203

[0.699, 0.982]

Sp

0.754 ± 0.305

[0.542, 0.966]

0.815 ± 0.157

[0.706, 0.924]

0.739 ± 0.230

[0.579, 0.899]

0.807 ± 0.137

[0.711, 0.902]

0.874 ± 0.247

[0.737, 1.000]

0.864 ± 0.239

[0.697, 1.000]

0.882 ± 0.171

[0.763, 1.000]

0.839 ± 0.335

[0.606, 1.000]

MCC

0.632 ± 0.395

[0.358, 0.907]

0.692 ± 0.215

[0.543, 0.841]

0.681 ± 0.160

[0.569, 0.793]

0.719 ± 0.191

[0.587, 0.852]

0.770 ± 0.427

[0.577, 1.000]

0.808 ± 0.295

[0.603, 1.000]

0.730 ± 0.212

[0.583, 0.878]

0.725 ± 0.319

[0.503, 0.947]

VGG19

Ac

0.724 ± 0.169

[0.570, 0.928]

0.847 ± 0.081

[0.791, 0.903]

0.799 ± 0.049

[0.765, 0.833]

0.837 ± 0.101

[0.767, 0.906]

0.796 ± 0.309

[0.754, 0.999]

0.880 ± 0.136

[0.785, 0.975]

0.857 ± 0.135

[0.763, 0.950]

0.903 ± 0.125

[0.816, 0.989]

Pe

0.801 ± 0.098

[0.757, 0.879]

0.868 ± 0.059

[0.827, 0.909]

0.838 ± 0.037

[0.812, 0.863]

0.847 ± 0.099

[0.777, 0.916]

0.904 ± 0.139

[0.789, 0.994]

0.895 ± 0.122

[0.810, 0.979]

0.874 ± 0.116

[0.795, 0.956]

0.911 ± 0.112

[0.833, 0.988]

F1

0.702 ± 0.209

[0.465, 0.968]

0.842 ± 0.086

[0.783, 0.902]

0.792 ± 0.056

[0.753, 0.831]

0.835 ± 0.102

[0.765, 0.906]

0.895 ± 0.144

[0.735, 1.000]

0.878 ± 0.139

[0.782, 0.975]

0.854 ± 0.139

[0.757, 0.951]

0.902 ± 0.125

[0.815, 0.989]

Sp

0.876 ± 0.346

[0.532, 1.000]

0.720 ± 0.223

[0.565, 0.875]

0.734 ± 0.333

[0.502, 0.965]

0.816 ± 0.149

[0.713, 0.919]

0.842 ± 0.185

[0.868, 0.992]

0.859 ± 0.223

[0.705, 1.000]

0.897 ± 0.204

[0.755, 1.000]

0.884 ± 0.228

[0.726, 1.000]

MCC

0.521 ± 0.271

[0.258, 0.829]

0.710 ± 0.142

[0.612, 0.809]

0.633 ± 0.072

[0.583, 0.683]

0.683 ± 0.200

[0.544, 0.822]

0.801 ± 0.282

[0.531, 0.993]

0.774 ± 0.258

[0.595, 0.953]

0.730 ± 0.252

[0.556, 0.906]

0.814 ± 0.236

[0.649, 0.977]

Xception

Ac

0.921 ± 0.160

[0.809, 1.000]

0.936 ± 0.140

[0.839, 1.000]

0.925 ± 0.168

[0.809, 1.000]

0.888 ± 0.171

[0.768, 1.000]

0.945 ± 0.126

[0.858, 1.000]

0.952 ± 0.109

[0.876, 1.000]

0.939 ± 0.137

[0.845, 1.000]

0.932 ± 0.154

[0.826, 1.000]

Pe

0.844 ± 0.119

[0.854, 1.000]

0.938 ± 0.137

[0.843, 1.000]

0.932 ± 0.159

[0.822, 1.000]

0.908 ± 0.136

[0.814, 1.000]

0.912 ± 0.115

[0.869, 1.000]

0.956 ± 0.098

[0.888, 1.000]

0.942 ± 0.133

[0.850, 1.000]

0.937 ± 0.142

[0.838, 1.000]

F1

0.918 ± 0.168

[0.801, 1.000]

0.936 ± 0.140

[0.839, 1.000]

0.925 ± 0.169

[0.806, 1.000]

0.885 ± 0.177

[0.761, 1.000]

0.945 ± 0.127

[0.857, 1.000]

0.952 ± 0.111

[0.875, 1.000]

0.939 ± 0.137

[0.844, 1.000]

0.931 ± 0.158

[0.821, 1.000]

Sp

0.873 ± 0.334

[0.612, 1.000]

0.945 ± 0.107

[0.871, 1.000]

0.880 ± 0.264

[0.696, 1.000]

0.840 ± 0.354

[0.594, 1.000]

0.912 ± 0.193

[0.780, 1.000]

0.937 ± 0.200

[0.798, 1.000]

0.966 ± 0.094

[0.901, 1.000]

0.968 ± 0.071

[0.919, 1.000]

MCC

0.851 ± 0.284

[0.659, 1.000]

0.874 ± 0.277

[0.682, 1.000]

0.857 ± 0.328

[0.629, 1.000]

0.796 ± 0.308

[0.581, 1.000]

0.895 ± 0.241

[0.727, 1.000]

0.908 ± 0.210

[0.762, 1.000]

0.882 ± 0.270

[0.694, 1.000]

0.868 ± 0.301

[0.659, 1.000]

Inception-ResNetV2

Ac

0.745 ± 0.177

[0.622, 0.868]

0.884 ± 0.156

[0.776, 0.992]

0.888 ± 0.116

[0.808, 0.969]

0.841 ± 0.124

[0.755, 0.927]

0.753 ± 0.150

[0.649, 0.857]

0.898 ± 0.138

[0.802, 0.993]

0.880 ± 0.145

[0.779, 0.980]

0.909 ± 0.125

[0.823, 0.997]

Pe

0.817 ± 0.069

[0.768, 0.865]

0.901 ± 0.129

[0.811, 0.990]

0.903 ± 0.088

[0.842, 0.964]

0.875 ± 0.083

[0.817, 0.932]

0.823 ± 0.067

[0.776, 0.870]

0.913 ± 0.102

[0.843, 0.984]

0.904 ± 0.101

[0.834, 0.974]

0.921 ± 0.103

[0.849, 0.992]

F1

0.722 ± 0.233

[0.560, 0.883]

0.880 ± 0.163

[0.767, 0.992]

0.886 ± 0.123

[0.801, 0.971]

0.836 ± 0.131

[0.745, 0.927]

0.734 ± 0.199

[0.595, 0.872]

0.895 ± 0.146

[0.794, 0.996]

0.877 ± 0.151

[0.772, 0.982]

0.909 ± 0.127

[0.821, 0.997]

Sp

0.688 ± 0.363

[0.436, 0.940]

0.786 ± 0.296

[0.580, 0.992]

0.824 ± 0.276

[0.632, 0.992]

0.710 ± 0.290

[0.509, 0.912]

0.723 ± 0.583

[0.318, 1.000]

0.860 ± 0.320

[0.638, 1.000]

0.843 ± 0.338

[0.608, 1.000]

0.872 ± 0.243

[0.704, 1.000]

MCC

0.551 ± 0.257

[0.436, 0.941]

0.781 ± 0.288

[0.581, 0.981]

0.790 ± 0.207

[0.646, 0.934]

0.714 ± 0.290

[0.571, 0.858]

0.568 ± 0.227

[0.410, 0.726]

0.809 ± 0.244

[0.640, 0.979]

0.784 ± 0.246

[0.613, 0.954]

0.831 ± 0.226

[0.673, 0.988]

DenseNet201

Ac

0.843 ± 0.110

[0.766, 0.920]

0.952 ± 0.081

[0.895, 1.000]

0.943 ± 0.131

[0.852, 1.000]

0.947 ± 0.139

[0.851, 1.000]

0.942 ± 0.073

[0.891, 0.992]

0.975 ± 0.085

[0.915, 1.000]

0.973 ± 0.074

[0.921, 1.000]

0.974 ± 0.087

[0.913, 1.000]

Pe

0.871 ± 0.080

[0.815, 0.926]

0.954 ± 0.073

[0.904, 1.000]

0.952 ± 0.106

[0.878, 1.000]

0.948 ± 0.137

[0.853, 1.000]

0.947 ± 0.058

[0.906, 0.987]

0.976 ± 0.079

[0.920, 1.000]

0.973 ± 0.072

[0.923, 1.000]

0.974 ± 0.087

[0.914, 1.000]

F1

0.839 ± 0.116

[0.758, 0.920]

0.951 ± 0.082

[0.894, 1.000]

0.942 ± 0.135

[0.848, 1.000]

0.948 ± 0.140

[0.850, 1.000]

0.941 ± 0.073

[0.890, 0.993]

0.974 ± 0.086

[0.914, 1.000]

0.972 ± 0.074

[0.921, 1.000]

0.974 ± 0.087

[0.914, 1.000]

Sp

0.807 ± 0.339

[0.571, 1.000]

0.925 ± 0.158

[0.816, 1.000]

0.890 ± 0.265

[0.706, 1.000]

0.942 ± 0.180

[0.817, 1.000]

0.921 ± 0.168

[0.804, 1.000]

0.955 ± 0.147

[0.853, 1.000]

0.980 ± 0.044

[0.949, 1.000]

0.975 ± 0.090

[0.912, 1.000]

MCC

0.712 ± 0.190

[0.580, 0.845]

0.906 ± 0.156

[0.797, 1.000]

0.895 ± 0.240

[0.728, 1.000]

0.896 ± 0.277

[0.704, 1.000]

0.889 ± 0.131

[0.798, 0.980]

0.950 ± 0.166

[0.835, 1.000]

0.946 ± 0.146

[0.844, 1.000]

0.948 ± 0.175

[0.827, 1.000]

MobileNetV2

Ac

0.796 ± 0.224

[0.640, 0.952]

0.965 ± 0.098

[0.896, 1.000]

0.961 ± 0.123

[0.875, 1.000]

0.947 ± 0.159

[0.836, 1.000]

0.952 ± 0.093

[0.888, 1.000]

0.981 ± 0.066

[0.935, 1.000]

0.971 ± 0.077

[0.917, 1.000]

0.961 ± 0.133

[0.869, 1.000]

Pe

0.851 ± 0.104

[0.778, 0.922]

0.965 ± 0.097

[0.897, 1.000]

0.963 ± 0.114

[0.884, 1.000]

0.958 ± 0.121

[0.874, 1.000]

0.955 ± 0.088

[0.894, 1.000]

0.981 ± 0.066

[0.935, 1.000]

0.972 ± 0.077

[0.918, 1.000]

0.968 ± 0.108

[0.893, 1.000]

F1

0.779 ± 0.276

[0.587, 0.971]

0.964 ± 0.098

[0.896, 1.000]

0.960 ± 0.125

[0.873, 1.000]

0.946 ± 0.165

[0.831, 1.000]

0.952 ± 0.093

[0.890, 1.000]

0.981 ± 0.066

[0.935, 1.000]

0.971 ± 0.077

[0.917, 1.000]

0.961 ± 0.133

[0.867, 1.000]

Sp

0.835 ± 0.269

[0.648, 1.000]

0.954 ± 0.135

[0.861, 1.000]

0.937 ± 0.207

[0.793, 1.000]

0.901 ± 0.309

[0.686, 1.000]

0.934 ± 0.150

[0.830, 1.000]

0.981 ± 0.070

[0.932, 1.000]

0.965 ± 0.089

[0.902, 1.000]

0.934 ± 0.231

[0.774, 1.000]

MCC

0.639 ± 0.342

[0.402, 0.877]

0.929 ± 0.197

[0.793, 1.000]

0.923 ± 0.239

[0.757, 1.000]

0.904 ± 0.282

[0.709, 1.000]

0.908 ± 0.180

[0.782, 1.000]

0.963 ± 0.133

[0.870, 1.000]

0.943 ± 0.154

[0.836, 1.000]

0.930 ± 0.240

[0.763, 1.000]

  1. The performance metrics—accuracy (Ac), weighted precision (Pe), weighted F1-score (F1), specificity (Sp), and Matthew’s Correlation coefficient (MCC) values recorded by six DL models in the classification of FCD-II using 2D images of three planes together (Iacs) and separately (axial, coronal, and sagittal) of two MRI modes—FLAIR and T1w have been displayed. PMV indicates performance metrics values and the numbers in bold fonts (black color) indicate the highest performance values recorded by respective models for the respective classification tasks. Iacs indicates images of all planes together, ax represents images of the axial plane, cr depicts images of the coronal plane, and sg indicates images of the sagittal plane. Results format: mean ± standard deviation [0.95 confidence interval (CI)]. The Xception model recorded the highest classification performance values for Iacs for both MRI modes, while DenseNet201 achieved maximum scores for the axial plane. Furthermore, the lightweight neural network, MobileNetV2 obtained the highest performance values for Iacs and the axial plane of T1w MRI mode.
Back to article page