Fully Automated Delineation of Gross Tumor Volume for Head and Neck Cancer on PET-CT Using Deep Learning: A Dual-Center Study
Table 2
The segmentation performance of all tumors by using the proposed method in Experiment 2.
Patient number
GTVm (cm3)
GTVa (cm3)
Sensitivity
Precision
DSC
MSD (mm)
Database 1
1
73.1
61.4
0.619
0.736
0.683
4.8
2
10.4
8.9
0.759
0.886
0.818
2.1
3
3.5
3.7
0.689
0.653
0.670
4.1
4
14.5
16.9
0.832
0.715
0.769
2.6
5
25.9
22.6
0.736
0.844
0.786
2.5
6
27.1
25.7
0.770
0.812
0.791
3.7
7
11.8
11.6
0.713
0.723
0.718
7.0
8
37.5
28.1
0.422
0.562
0.482
10.3
9
65.7
69.5
0.841
0.794
0.817
2.9
10
13.3
12.1
0.534
0.588
0.560
11.5
11
26.7
26.3
0.787
0.799
0.793
1.8
12
17.6
13.5
0.601
0.783
0.680
8.0
13
44.0
44.5
0.873
0.863
0.868
1.2
14
51.8
48.2
0.788
0.847
0.817
2.5
15
73.1
58.2
0.729
0.915
0.812
2.5
16
147.5
147.1
0.782
0.784
0.783
2.5
17
12.1
14.4
0.820
0.692
0.750
2.6
Mean ± SD
—
—
0.723 ± 0.119
0.764 ± 0.100
0.741 ± 0.100
4.2 ± 3.1
Database 2
1
24.5
30.7
0.594
0.475
0.528
13.1
2
80.4
74.0
0.791
0.859
0.824
1.7
3
18.3
20.6
0.918
0.815
0.863
2.9
4
18.2
17.3
0.772
0.815
0.793
6.7
5
69.8
43.0
0.478
0.775
0.591
6.7
Mean ± SD
—
—
0.711 ± 0.174
0.748 ± 0.155
0.720 ± 0.150
6.2 ± 4.5
Note: GTVm, the gross tumor volume of manual delineation; GTVa, the gross tumor volume of automatic segmentation by the proposed method; DSC, dice similarity coefficient; MSD, mean surface distance.