Skip to main content

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

From: Clinical implementation of MRI-based organs-at-risk auto-segmentation with convolutional networks for prostate radiotherapy

StudyPtsModalityMethod(s)BladderRectumFemurLFemurR
    DSCDSCDSCDSC
    HD95HD95HD95HD95
    MSDMSDMSDMSD
Convolutional network-based
Men2017 [53]218/60 CT2D0.92 0.930.92
   dilated 
   VGG-16 
Feng2018 [27]30/10 MRIMulti-task0.952 ±0.0070.88 ±0.03 
   residual 
   2D FCN 
Kazemifar2018 [54]51/9/20 CT2D0.95 ±0.040.92±0.06  
   U-net0.4±0.60.2±0.3 
    1.1 ±0.8a0.8±0.6a 
Balagopal2018 [55]108/28CT2D U-net0.95 ±0.020.84 ±0.040.96 ±0.030.95 ±0.01
 mean + 3D U-net17.0 ±14.64.9 ±3.9 
 4 models (ResNeXT)0.5 ±0.70.8 ±0.7 
Dong2019 [56]140x5 +MRI3D Cycle-GAN0.95 ±0.030.89 ±0.04 
   + deep attention6.81 ±9.2510.84 ±15.59 
   U-net0.52±0.220.92 ±1.03 
Elguindi2019 [49]40/10/50MRI 0.93 ±0.040.82 ±0.05 
   DeepLabV3+ 
    0.92 ±0.1b0.87 ±0.07b 
This study97/53 MRI3D0.96±0.020.88 ±0.050.97±0.010.97±0.01
   multi-scale2.5 ±1.17.4 ±4.41.6±0.51.5±0.5
   DeepMedic0.6±0.31.7 ±0.80.5±0.10.5±0.1
    0.98 ±0.03c0.92 ±0.05c0.989±0.008c0.997±0.003c
Conventional
LaMacchia2012 [16]5CTABAS 2.00.93 ±0.030.77 ±0.070.94 ±0.040.94 ±0.04
   VelocityAI 2.6.20.72 ±0.150.75 ±0.040.92 ±0.020.92 ±0.03
   MIM 5.1.10.93 ±0.020.87 ±0.050.94 ±0.020.94 ±0.01
Dowling2015 [17]39MRImulti-atlas0.86 ±0.120.84 ±0.060.91 ±0.03
   voting    
   diffeomorphic reg5.1 ±4.62.4 ±1.01.5 ±0.5
Delpon2016 [52]10/10 CTMirada0.76 ±0.120.73 ±0.070.89 ±0.050.91 ±0.03
    15 ±910 ±30.2 ±6.48.1 ±5.6
   MIM0.80 ±0.140.75 ±0.070.89 ±0.080.92 ±0.02
    14.0 ±6.39.9 ±3.49.9 ±7.98.2 ±5.3
   ABAS0.81 ±0.130.75 ±0.090.91 ±0.040.92 ±0.02
    13.6 ±7.99.9 ±4.48.6 ±6.98.5 ±6.1
   SPICE0.76 ±0.260.68 ±0.120.70 ±0.050.72 ±0.03
    9.2 ±11.713 ±529.7 ±9.030 ±6.5
   Raystation0.59 ±0.150.49 ±0.120.91 ±0.030.92 ±0.02
    28.5 ±13.116.5 ±3.78.8 ±7.26.4 ±5.0
  1. training/(validation)/test; + indicating x... cross-fold validation; a mean surface Hausdorff distance; b,c surface dice similarity coefficient as in [48] with τ=3 or 2 mm, respectively