Table 3 Overview of the performance of automatic OARs delineations based on MRI and CT subdivided in convolutional network-based and conventional approaches. The number of patients included in the study (Pts), the imaging modality, a brief description of the method and metrics as dice similarity coefficient (DSC), 95% boundary Hausdorff distance (HD₉₅) and mean surface distance (MSD) were reported for each study. HD₉₅ and MSD are expressed in mm

Convolutional network-based

Men2017 [53]

218/60 ^∗

0.92

0.93

0.92

Feng2018 [27]

30/10 ^∗

0.952 ±0.007

0.88 ±0.03

Kazemifar2018 [54]

51/9/20 ^∗

0.95 ±0.04

0.92±0.06

0.4±0.6

0.2±0.3

1.1 ±0.8^a

0.8±0.6^a

Balagopal2018 [55]

108/28

0.95 ±0.02

0.84 ±0.04

0.96 ±0.03

0.95 ±0.01

mean

17.0 ±14.6

4.9 ±3.9

4 models

0.5 ±0.7

0.8 ±0.7

Dong2019 [56]

140x5 ⁺

0.95 ±0.03

0.89 ±0.04

6.81 ±9.25

10.84 ±15.59

0.52±0.22

0.92 ±1.03

Elguindi2019 [49]

40/10/50

0.93 ±0.04

0.82 ±0.05

0.92 ±0.1^b

0.87 ±0.07^b

This study

97/53 ^∗

0.96±0.02

0.88 ±0.05

0.97±0.01

0.97±0.01

2.5 ±1.1

7.4 ±4.4

1.6±0.5

1.5±0.5

0.6±0.3

1.7 ±0.8

0.5±0.1

0.5±0.1

0.98 ±0.03^c

0.92 ±0.05^c

0.989±0.008^c

0.997±0.003^c

Conventional

LaMacchia2012 [16]

5

0.93 ±0.03

0.77 ±0.07

0.94 ±0.04

0.94 ±0.04

0.72 ±0.15

0.75 ±0.04

0.92 ±0.02

0.92 ±0.03

0.93 ±0.02

0.87 ±0.05

0.94 ±0.02

0.94 ±0.01

Dowling2015 [17]

39

0.86 ±0.12

0.84 ±0.06

5.1 ±4.6

2.4 ±1.0

1.5 ±0.5

Delpon2016 [52]

10/10 ^∗

0.76 ±0.12

0.73 ±0.07

0.89 ±0.05

0.91 ±0.03

15 ±9

10 ±3

0.2 ±6.4

8.1 ±5.6

0.80 ±0.14

0.75 ±0.07

0.89 ±0.08

0.92 ±0.02

14.0 ±6.3

9.9 ±3.4

9.9 ±7.9

8.2 ±5.3

0.81 ±0.13

0.75 ±0.09

0.91 ±0.04

0.92 ±0.02

13.6 ±7.9

9.9 ±4.4

8.6 ±6.9

8.5 ±6.1

0.76 ±0.26

0.68 ±0.12

0.70 ±0.05

0.72 ±0.03

9.2 ±11.7

13 ±5

29.7 ±9.0

30 ±6.5

0.59 ±0.15

0.49 ±0.12

0.91 ±0.03

0.92 ±0.02

28.5 ±13.1

16.5 ±3.7

8.8 ±7.2

6.4 ±5.0