Skip to main content

Table 5 Summary of metric selection guidelines

From: Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool

 

DICE

JAC

TPR

TNR

FPR

FNR

FMS

VS

GCE

RI

ARI

MI

VOI

ICC

PBD

KAP

AUC

HD

AVD

MHD

Outliers exist

✓

✓

    

✓

✓

   

✓

✓

  

✓

✓

X

✓

✓

Small segment

X

X

X

X

X

X

X

  

X

X

X

X

  

X

X

✓

✓

✓

Complex boundary

       

X

         

✓

✓

X

Low densities

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

✓

✓

✓

Low segmentation quality

       

X

         

✓

✓

✓

Contour is important

       

X

         

✓

✓

X

Alignment is important

       

X

            

Recall is important

  

✓

        

✓

        

Volume is important

       

✓

            

General shape & alignment

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

✓

  1. Each row corresponds to either a segmentation property or a requirement and each column corresponds to one of the metrics in Table 1. A checked cell (\(\checkmark \)) denotes that the metric is recommended for the corresponding property/requirement, a crossed cell (X) denotes that the metric is not recommended, and empty cells denote neutrality