Multilevel Strip Pooling-Based Convolutional Neural Network for the Classification of Carotid Plaque Echogenicity
Table 3
Sensitivity and specificity comparisons among VGG16, SPP-based VGGNet, and MSP-based VGGNet.
Fold
Methods
Performance evaluation (%)
1
VGG16
82.6
73.3
92.6
82.8
96.5
92.9
83.1
90.8
SPP-VGG
95.7
91.1
87.4
91.4
97.6
92.5
94.9
95.0
MSP-VGG
98.6
91.1
89.6
93.1
98.1
94.5
94.9
95.8
2
VGG16
84.6
67.5
92.8
81.6
98.0
91.2
82.3
90.5
SPP-VGG
98.5
87.0
93.4
93.0
97.7
96.3
94.3
96.1
MSP-VGG
98.5
90.9
92.8
94.1
98.1
95.8
95.7
96.6
3
VGG16
70.8
66.7
94.3
77.3
98.9
88.5
77.8
88.4
SPP-VGG
91.7
76.5
90.1
86.1
97.9
92.8
86.5
92.4
MSP-VGG
94.4
85.2
92.9
90.9
99.5
94.8
90.7
95.0
4
VGG16
78.8
64.9
96.0
79.9
99.0
92.9
77.3
89.7
SPP-VGG
93.9
81.8
96.0
90.6
97.6
97.2
90.6
95.1
MSP-VGG
95.5
81.8
96.0
91.1
98.6
96.3
91.3
95.4
5
VGG16
73.1
63.4
89.0
75.2
99.5
84.8
76.5
86.9
SPP-VGG
94.0
85.4
89.0
89.5
97.6
92.8
91.7
94.0
MSP-VGG
92.5
89.0
92.4
91.3
98.6
93.8
93.8
95.4
VGG16
SPP-VGG
MSP-VGG
SPP-VGG and MSP-VGG are short for SPP-based VGGNet and MSP-based VGGNet, respectively. The best results are highlighted in bold. The listed metrics were obtained on the test dataset.