Conformity Metrics¶

Conformity metrics describe how the prescription isodose overlaps a target. All conformity.compute_* functions are reference-free: they characterize one dose distribution. The Paddick conformity-index distance is reference-based and compares the single-plan index between two doses.

Let:

\(V_{\mathrm{target}}\) be the target volume;
\(V_{\mathrm{Rx}}\) be the volume receiving at least the prescription dose;
\(V_{\mathrm{target,Rx}}\) be their intersection.

Classification¶

Metric	API	Reference use	Ideal or direction
Coverage	`conformity.compute_coverage`	Reference-free	1
Spillage	`conformity.compute_spillage`	Reference-free	0
Conformity index	`conformity.compute_conformity_index`	Reference-free	1
Conformation number	`conformity.compute_conformity_number`	Reference-free	1
Paddick conformity index	`conformity.compute_paddick_conformity_index`	Reference-free	1
RTOG conformity index	`conformity.compute_rtog_conformity_index`	Reference-free	1
Prescription-dose MAE	`conformity.compute_prescription_mae`	Reference-free	Lower
Paddick CI distance	`compare_paddick_conformity_index`	Reference-based	Lower

Coverage¶

Coverage is the fraction of target voxels receiving at least the prescription dose:

\[ \mathrm{Coverage} =\frac{V_{\mathrm{target,Rx}}}{V_{\mathrm{target}}}. \]

from dosemetrics.metrics import conformity

coverage = conformity.compute_coverage(dose, ptv, prescription_dose=60.0)

Spillage¶

Spillage is the fraction of the prescription-isodose volume outside the target:

\[ \mathrm{Spillage} =\frac{V_{\mathrm{Rx}}-V_{\mathrm{target,Rx}}}{V_{\mathrm{Rx}}}. \]

spillage = conformity.compute_spillage(dose, ptv, prescription_dose=60.0)

When \(V_{\mathrm{Rx}}>0\), compute_spillage is the complement of compute_conformity_index for the same dose, target, and prescription. Both functions return 0 when no voxel reaches the prescription dose.

Conformity index¶

This implementation reports prescription-isodose purity:

\[ \mathrm{CI} =\frac{V_{\mathrm{target,Rx}}}{V_{\mathrm{Rx}}}. \]

ci = conformity.compute_conformity_index(dose, ptv, prescription_dose=60.0)

Conformation number and Paddick CI¶

Both implemented functions combine coverage and prescription-isodose purity:

\[ \mathrm{CN}=\mathrm{CI}_{\mathrm{Paddick}} =\frac{V_{\mathrm{target,Rx}}}{V_{\mathrm{target}}} \frac{V_{\mathrm{target,Rx}}}{V_{\mathrm{Rx}}} =\frac{V_{\mathrm{target,Rx}}^2} {V_{\mathrm{target}}V_{\mathrm{Rx}}}. \]

cn = conformity.compute_conformity_number(dose, ptv, prescription_dose=60.0)
paddick_ci = conformity.compute_paddick_conformity_index(
    dose, ptv, prescription_dose=60.0
)

Paddick conformity-index distance¶

Reference-based · compare_paddick_conformity_index · dimensionless · lower is better.

Paddick conformity-index distance The same Paddick CI is computed independently for each plan before taking the absolute difference.

\[ \mathrm{PCID} =\left|\mathrm{CI}_{\mathrm{evaluated}} -\mathrm{CI}_{\mathrm{reference}}\right|. \]

from dosemetrics.metrics import compare_paddick_conformity_index

pcid = compare_paddick_conformity_index(
    reference,
    evaluated,
    ptv_high,
    prescription_dose=70.0,
)

The direct single-plan version is conformity.compute_paddick_conformity_index.

RTOG conformity index¶

The RTOG size ratio compares the complete prescription-isodose volume with the target volume:

\[ \mathrm{CI}_{\mathrm{RTOG}} =\frac{V_{\mathrm{Rx}}}{V_{\mathrm{target}}}. \]

rtog_ci = conformity.compute_rtog_conformity_index(
    dose, ptv, prescription_dose=60.0
)

Unlike overlap-aware indices, this ratio alone does not distinguish correctly placed prescription dose from an equal-sized but displaced isodose volume.

Prescription-dose MAE¶

Prescription-dose MAE measures voxel-wise deviation from the prescribed dose inside the target:

\[ \mathrm{MAE}_{\mathrm{Rx}} =\frac{1}{\left|V_{\mathrm{target}}\right|} \sum_{v\in V_{\mathrm{target}}} \left|D(v)-D_{\mathrm{Rx}}\right|. \]

mae_gy = conformity.compute_prescription_mae(
    dose, ptv, prescription_dose=60.0
)

This single-plan quantity is not the same as PTV mean-dose distance, which compares two plans.