Table 2 Segmentation metrics.

From: Convolutional neural networks for the automatic segmentation of lumbar paraspinal muscles in people with low back pain

Metric

Equation

Range

Meaning

Sørensen-Dice Index

\(\frac{2\times \left|\hbox{SM}\cap \hbox{GT}\right|}{|\hbox{SM}| + |\hbox{GT}|}\)

0–1

Spatial overlap between masks

Jaccard Index

\(\frac{|\hbox{SM}\cap \hbox{GT}|}{|\hbox{SM}| + |\hbox{GT}| - |\hbox{SM }\cap \hbox{GT}|}\)

0–1

Spatial overlap between masks

Conformity Coefficient

\(1- \frac{\hbox{FP }+\hbox{FN}}{\hbox{TP}}\)

 < 1

Ratio of incorrectly and correctly segmented voxels

True Positive Rate (TPR)

\(\frac{\hbox{TP}}{\hbox{TP }+\hbox{FN}}\)

0–1

Sensitivity

True Negative Rate (TNR)

\(\frac{\hbox{TN}}{\hbox{TN }+\hbox{FP}}\)

0–1

Specificity

Positive Predictive Value (PPV)

\(\frac{\hbox{TP}}{\hbox{TP }+\hbox{FP}}\)

0–1

Precision

Volume Ratio

\(\frac{\hbox{SM}}{\hbox{GT}}\)

 ≥ 0

Ratio of mask volumes

  1. SM  segmentation mask, GT  ground truth mask, TP  true positive (i.e., voxels correctly segmented as muscle), TN  true negative (i.e., voxels correctly segmented as background), FP  false positive (i.e., voxels incorrectly segmented as muscle), FN  false negative (i.e., voxels incorrectly segmented as background).
Back to article page