Hybrid transformer convolutional neural network-based radiomics models for osteoporosis screening in routine CT

BMC Medical Imaging

Table 2 Performance of automatic segmentation

Vertebral body	Interrater(n = 5)	Test1 (n = 35)		Test2 (n = 15)
Vertebral body	DSCs	DSCs	ASD	DSCs	ASD
L1	0.965 ± 0.015	0.964 ± 0.012	0.605 ± 0.521	0.965 ± 0.05	0.452 ± 0.069
L2	0.969 ± 0.009	0.956 ± 0.048	0.772 ± 1.953	0.967 ± 0.05	0.432 ± 0.086
L3	0.967 ± 0.016	0.969 ± 0.008	0.367 ± 0.165	0.965 ± 0.010	0.467 ± 0.142
L4	0.943 ± 0.060	0.973 ± 0.010	0.336 ± 0.139	0.976 ± 0.009	0.422 ± 0.196
Trabecular compartment of the vertebral body
L1	0.962 ± 0.010	0.966 ± 0.022	0.431 ± 0.443	0.960 ± 0.012	0.539 ± 0.552
L2	0.968 ± 0.011	0.966 ± 0.024	0.445 ± 0.486	0.959 ± 0.017	0.706 ± 0.911
L3	0.968 ± 0.008	0.963 ± 0.036	0.387 ± 0.266	0.959 ± 0.012	0.511 ± 0.604
L4	0.966 ± 0.010	0.962 ± 0.049	0.430 ± 0.385	0.955 ± 0.032	0.507 ± 0.496

Mean dice similarity coefficient and average surface distance (± standard deviation) are used to evaluate the performance of the automatic segmentation framework
DSCs, dice similarity coefficient; ASD, average surface distance

ISSN: 1471-2342