# Plan Evaluation Metrics: DVH, Conformity Index, and Radiobiological Models

## Overview

Evaluating a radiation treatment plan involves integrating multiple quantitative metrics to assess its quality before delivery. No single metric can fully characterize a plan; therefore, a comprehensive evaluation requires analysis of dose-volume histograms (DVHs), conformity and homogeneity indices, and clinical judgment. Radiobiological models such as tumor control probability (TCP) and normal tissue complication probability (NTCP) add an additional layer of assessment by estimating clinical outcomes based on dosimetric data. This evaluation process is collaborative, involving the radiation oncologist, medical physicist, and dosimetrist to ensure the plan meets clinical goals.

## Dose-Volume Histogram (DVH)

### Cumulative DVH

The cumulative DVH is the most commonly used tool for plan evaluation. It plots the percentage or absolute volume of a structure receiving at least a given dose, with dose on the x-axis and volume (expressed as a percentage or cubic centimeters) on the y-axis. For a target volume, the ideal cumulative DVH curve shows a steep drop-off near the prescription dose, indicating a homogeneous dose delivery. In contrast, for an organ at risk (OAR), the curve should fall off as quickly as possible, reflecting minimal volume receiving high doses.

### Differential DVH

The differential DVH presents the volume receiving each specific dose level in a histogram or bin format. For a homogeneously irradiated target, this appears as a narrow spike at the prescription dose. Although less commonly used in clinical practice, differential DVHs are informative for identifying hot or cold spots within the dose distribution.

### Key DVH Parameters

Several key parameters extracted from DVHs guide plan evaluation. Dx% denotes the minimum dose to the hottest x% of the volume; for example, D95% represents the minimum dose received by 95% of the planning target volume (PTV) and is commonly used to assess prescription coverage. Vx Gy indicates the percentage of volume receiving at least x Gy, such as V20 for lung or V50 for rectum. Dmax, often reported as D0.03cc, represents the maximum point dose to the hottest 0.03 cubic centimeters to avoid artifacts from single voxels. Dmean is the mean dose to the structure and is particularly relevant for parallel organs like the lung, parotid, heart, and liver. D2% is the near-maximum dose, defined as the dose to the hottest 2% of the volume, preferred over a single-point maximum for robustness. D98% is the near-minimum dose covering 98% of the volume and reflects the adequacy of target coverage.

### DVH Limitations

Despite their utility, DVHs have limitations. They reduce three-dimensional spatial dose information into a two-dimensional curve, meaning two very different dose distributions can produce identical DVHs. They do not convey the spatial location of high or low doses within the structure, nor do they account for functional heterogeneity within tissues, assuming uniform importance throughout the volume. Therefore, DVH analysis must always be complemented by visual review of isodose distributions on CT slices.

## Conformity Metrics

### Conformity Index (CI)

The conformity index measures how well the prescription isodose conforms to the target volume. Several definitions exist.

#### RTOG Conformity Index

The RTOG conformity index is calculated as the ratio of the volume of the reference (prescription) isodose (V_RI) to the target volume (V_TV). An ideal CI equals 1.0, indicating perfect conformity. A CI greater than 1 suggests that normal tissue beyond the target is being treated, while a CI less than 1 indicates undercoverage of the target.

#### Paddick Conformity Index

More commonly used in stereotactic radiosurgery (SRS) and stereotactic body radiation therapy (SBRT), the Paddick conformity index accounts for both conformity and coverage simultaneously. It is defined as the square of the volume of the target within the prescription isodose (TV_PIV) divided by the product of the target volume (TV) and the prescription isodose volume (PIV). An ideal value is 1.0, with values less than 1 indicating either undercoverage, poor conformity, or both.

### Gradient Index (GI)

The gradient index quantifies how rapidly the dose falls off outside the target. It is calculated as the ratio of the volume of the 50% isodose to the prescription isodose volume (V_50% / V_100%). A lower GI indicates a steeper dose falloff and better sparing of normal tissue beyond the target. This metric is particularly important in SRS and SBRT, where rapid dose falloff is critical.

### Homogeneity Index (HI)

The homogeneity index measures dose uniformity within the target and is calculated as (D2% - D98%) divided by D50%. An ideal HI is zero, indicating perfectly uniform dose. Higher values reflect greater dose heterogeneity. Acceptable HI values depend on the treatment technique and clinical context: for three-dimensional conformal radiation therapy (3D-CRT), HI should be less than 0.10 to 0.15; for intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT), less than 0.10 is typical. In SRS and SBRT, higher heterogeneity is often acceptable, as hotspots within the target may be beneficial.

### ICRU Reference Point

The International Commission on Radiation Units and Measurements (ICRU) recommends reporting dose at a reference point, typically the isocenter or a point representative of the target dose. For IMRT, ICRU Report 83 advises reporting the median dose (D50%), near-maximum dose (D2%), and near-minimum dose (D98%).

## Radiobiological Models

### Tumor Control Probability (TCP)

Tumor control probability estimates the likelihood of local tumor control based on the dose distribution. It is derived from cell survival models using Poisson statistics applied to clonogenic cell kill. The TCP is expressed as exp(-N * SF), where N is the number of clonogenic cells and SF is the surviving fraction at the delivered dose. Clinically, TCP curves are sigmoid, showing a steep dose-response relationship near the dose required for tumor control. TCP is more useful for comparing plans than for absolute outcome prediction due to uncertainties in model parameters.

### Normal Tissue Complication Probability (NTCP)

Normal tissue complication probability estimates the risk of a specific complication based on the dose-volume distribution to an organ at risk. The most common model is the Lyman-Kutcher-Burman (LKB) model, which uses parameters such as TD50 (dose causing 50% complication probability), m (slope of the dose-response curve), and n (volume effect parameter). An n value of 1 indicates a strong volume effect typical of parallel organs where mean dose drives toxicity (e.g., lung pneumonitis). An n approaching zero indicates a weak volume effect typical of serial organs where maximum dose drives toxicity (e.g., spinal cord myelopathy). Logistic regression models also exist, fitting complication rates statistically to dose-volume parameters from clinical datasets.

### Equivalent Uniform Dose (EUD)

Equivalent uniform dose converts a heterogeneous dose distribution into a single uniform dose that would produce the same biological effect. The generalized EUD (gEUD) is calculated as the sum over dose bins of the volume fraction times the dose raised to the power a, all raised to the 1/a power. The parameter a is tissue-specific: a = 1 corresponds to mean dose (pure parallel organ), a approaching infinity corresponds to maximum dose (pure serial organ), and a approaching negative infinity corresponds to minimum dose (tumor). Negative a values model tumors where cold spots dominate, while positive a values model normal tissues where hotspots increase risk. EUD is useful for optimization objectives and plan comparison.

### Biologically Effective Dose (BED) and EQD2

The biologically effective dose accounts for fraction size effects and is calculated as BED = nd(1 + d/[α/β]), where n is the number of fractions, d is dose per fraction, and α/β is the tissue-specific ratio. EQD2 converts BED to an equivalent dose in 2 Gy fractions using EQD2 = BED / (1 + 2/[α/β]). These metrics are essential for comparing different fractionation schedules and require specification of the α/β ratio, typically 10 Gy for tumor acute effects and 3 Gy for late normal tissue effects.

## Practical Plan Evaluation Workflow

A systematic review of a treatment plan begins with assessing target coverage, ensuring that D95% meets or exceeds the prescription dose, V95% covers 95-98% of the PTV, and D98% (near-minimum dose) is adequate. Dose homogeneity is evaluated by confirming that D2% and D50% are acceptable and that no clinically significant hotspots exist outside the target. All organ-at-risk constraints specified by the protocol must be met, including both maximum doses for serial organs and mean or volume-based doses for parallel organs. Conformity indices should be appropriate for the treatment type, and the gradient index must be acceptable, especially for SBRT. Visual inspection of isodose lines on axial, coronal, and sagittal CT slices is essential to verify the spatial dose distribution. Low-dose regions, such as V5 and V10, should be checked for critical structures like lung, liver, and bone marrow. Plan deliverability is assessed by confirming that monitor units (MU) are reasonable and that complexity metrics fall within machine capabilities. Finally, the plan must align with the clinical context, including treatment intent and protocol requirements.

<image>A dose-volume histogram (DVH) for a head and neck IMRT plan showing curves for six structures: PTV-70 (red, steep drop at 70 Gy), PTV-56 (orange, steep drop at 56 Gy), left parotid (blue, with D50% marked), right parotid (green, with D50% marked), spinal cord (purple, with Dmax marked), and brainstem (pink, with Dmax marked). Key DVH metrics (D95, D2, Dmean, Dmax) are annotated on each curve with values. QUANTEC constraints are shown as dashed horizontal lines on the OAR curves for reference.</image>

<image>A visual comparison of two lung SBRT plans for the same target: Plan A has a conformity index of 0.85 (acceptable) but a gradient index of 5.0 (poor, with extensive low-dose spread), while Plan B has a conformity index of 0.90 and a gradient index of 3.2 (tight dose falloff). Axial dose color wash shows the prescription isodose and the 50% isodose volume for each plan, with the tighter 50% isodose in Plan B clearly visible. The gradient index formula and values are annotated.</image>

<image>A sigmoid curve graph showing tumor control probability (TCP) and normal tissue complication probability (NTCP) plotted against dose. The TCP curve rises steeply at the clinically relevant dose range (e.g., 60-80 Gy). The NTCP curve rises at a higher dose. The therapeutic window -- the dose range where TCP is high and NTCP is low -- is shaded in green. An arrow shows how IMRT shifts the NTCP curve to the right (by reducing OAR dose), widening the therapeutic window. The uncomplicated tumor control probability (UTCP = TCP x [1-NTCP]) curve is overlaid.</image>

## Key Clinical Pearls

Plan evaluation should never rely solely on DVH analysis; always review the spatial dose distribution on CT slices because a plan with perfect DVH metrics can still have clinically unacceptable dose distributions, such as hotspots in inappropriate locations. D95% coverage of the PTV is the standard prescription metric in most protocols, but D98%, which reflects the near-minimum dose, provides more information about the coldest part of the target and should be reported according to ICRU 83. For SBRT and SRS, conformity and gradient indices are the dominant quality metrics, while homogeneity is deliberately relaxed since hotspots within the target are accepted and may even be beneficial. When comparing plans with different fractionation schedules, always convert doses to biologically effective dose (BED) or equivalent dose in 2 Gy fractions (EQD2), as comparing physical doses directly is meaningless when fraction sizes differ. NTCP models are valuable for plan comparison and clinical decision-making, such as model-based selection for proton therapy, but they should never be treated as precise predictions due to inherent uncertainties from model parameters, patient variability, and data limitations. Ultimately, the best plan is not necessarily the one with the lowest organ-at-risk doses; it is the one that achieves adequate target coverage while maintaining complication risks at acceptable levels given the clinical context, including treatment intent, life expectancy, and comorbidities.

## References
- ICRU Report 83. "Prescribing, Recording, and Reporting Photon-Beam Intensity-Modulated Radiation Therapy (IMRT)." 2010.
- Paddick I. "A simple scoring ratio to index the conformity of radiosurgical treatment plans." *J Neurosurg*. 2000;93 Suppl 3:219-222.
- Niemierko A. "Reporting and analyzing dose distributions: a concept of equivalent uniform dose." *Med Phys*. 1997;24(1):103-110.
- Lyman JT. "Complication probability as assessed from dose-volume histograms." *Radiat Res Suppl*. 1985;8:S13-S19.
- Baumann M et al. "Exploring the role of radiobiological modelling in clinical practice." *Radiother Oncol*. 2012.