Metric availability by backend

Metric configuration keys are backend-specific. Each cell contains the exact key to place under metrics in the YAML file. An em dash means that the backend does not expose that metric.

Metric	AUDIT	pymia	Metrics Reloaded
Dice coefficient	dice	dice	dsc
Jaccard / Intersection over Union	jaccard	jacc	iou
Accuracy	accuracy	accu	accuracy
Precision / Positive Predictive Value	precision	prec	ppv
Sensitivity / Recall	sensitivity	sens	sensitivity
Specificity	specificity	spec	specificity
Hausdorff distance (maximum)	hausdorff_distance	haus	hd
Predicted physical area/volume	predicted_size	pred_vol	—
Reference physical area/volume	—	ref_vol	—
Number of predicted pixels/voxels	—	—	numb_pred
Number of reference pixels/voxels	—	—	numb_ref
True positives	—	tp	numb_tp
False positives	—	fp	numb_fp
False negatives	—	fn	numb_fn
True negatives	—	tn	—
F-measure / F-beta	—	fmeas	fbeta
Cohen's kappa	—	ckc	cohens_kappa
Adjusted Rand Index	—	ari	—
Area Under Curve	—	auc	—
Interclass Correlation	—	ic	—
Mutual Information	—	mi	—
Rand Index	—	rand	—
Surface Dice Overlap	—	sdo	—
Surface Overlap	—	so	—
Volume Similarity	—	vs	—
Average Distance	—	avd	—
Mahalanobis Distance	—	mahal	—
Variation of Information	—	vi	—
Global Consistency Error	—	gce	—
Probabilistic Distance	—	prob	—
Fallout	—	fallout	—
False Negative Rate	—	fnr	—
Balanced Accuracy	—	—	ba
Weighted Cohen's Kappa	—	—	wck
Positive Likelihood Ratio	—	—	lr+
Youden Index	—	—	youden_ind
Negative Predictive Value	—	—	npv
Intersection over Reference	—	—	ior
Matthews Correlation Coefficient	—	—	mcc
Net Benefit	—	—	nb
Normalised Expected Cost	—	—	ec
Centreline Dice	—	—	cldice
Average Symmetric Surface Distance	—	—	assd
Mean Average Surface Distance	—	—	masd
Hausdorff Distance Percentile	—	—	hd_perc
Normalized Surface Distance	—	—	nsd
Boundary IoU	—	—	boundary_iou
Absolute Volume Difference Ratio	—	—	avdr

predicted_size and pred_vol include image spacing. numb_pred is a raw pixel/voxel count, so these values coincide only when the spacing product is one.

pymia's avd, Metrics Reloaded's assd, and masd are shown on separate rows because their surface-distance definitions are not interchangeable.

Metrics Reloaded also implements probability, calibration, detection, and instance-segmentation metrics. They are not marked as available because AUDIT's current metric pipeline accepts discrete segmentation masks, not the probability maps or instance assignments those metric families require.