Quantitative and qualitative analysis¶
Two distinct approaches exist for reducing DCE curves to numerical measures, and they address different questions. The first characterizes the shape of the enhancement curve. The second estimates the physiological parameters that produced it.
Semi-quantitative analysis¶
Semi-quantitative measures are computed directly from the signal intensity curve, without conversion to concentration and without a kinetic model.
| Measure | Definition |
|---|---|
| Peak enhancement | Maximum signal increase above baseline, conventionally expressed as a percentage |
| Wash-in slope | Gradient of the initial upslope |
| Time to peak | Interval from arrival to maximum signal intensity |
| Washout | Behavior after the peak, classified as declining, plateau or persistent |
| IAUC | Initial area under the curve over a defined interval, commonly IAUC60 or IAUC90 |
| Curve type | The type I, II and III classification of persistent, plateau and washout morphology |
Strengths. No arterial input function, native T1 map or kinetic model is required, which eliminates the principal failure modes of quantitative analysis. The measures tolerate coarse temporal sampling, are computationally inexpensive, and are reproducible within an institution. A substantial proportion of the clinical literature, and several established reporting conventions, rest on semi-quantitative rather than pharmacokinetic measures; the curve-type classification used in breast MRI is one such example.
Limitations. The measures are expressed in arbitrary units. Percentage enhancement depends upon scanner, pulse sequence, flip angle, dose per unit body mass, injection rate and the native T1 of the tissue. Agreement between institutions examining the same subject is accordingly poor, as is agreement within an institution across a hardware or software change. The measures are moreover descriptive rather than mechanistic: an increased wash-in slope establishes that enhancement occurred more rapidly, but does not distinguish increased perfusion from increased permeability.
Pharmacokinetic modeling¶
Quantitative analysis converts signal intensity to gadolinium concentration and fits a model of contrast exchange to the resulting curves, with reference to the measured arterial input function.
The estimated parameters are physical quantities:
| Parameter | Definition | Units |
|---|---|---|
| Ktrans | Volume transfer constant between plasma and the extravascular extracellular space | min−1 |
| ve | Extravascular extracellular volume fraction | dimensionless, 0–1 |
| vp | Plasma volume fraction | dimensionless, 0–1 |
| kep | Efflux rate constant, equal to Ktrans/ve | min−1 |
Model selection depends upon the tissue under examination and upon what the acquisition can support. ROCKETSHIP implements the Tofts, extended Tofts, Patlak, two-compartment exchange, fast exchange regime and tissue uptake models. For a discussion of model differences and appropriate model selection see pharmacokinetic models.
Strengths. The estimated parameters retain meaning outside the dataset in which they were measured. Ktrans is in principle comparable across scanners, institutions and time points, which is the precondition for multi-center trials and for longitudinal assessment of treatment response. Modeling further resolves effects that curve morphology confounds: a lesion enhancing rapidly by virtue of perfusion and one enhancing rapidly by virtue of permeability are similar semi-quantitatively and distinguishable after fitting.
Limitations. The number of failure modes is considerably greater, and most are not evident in the output.
- An arterial input function is required, and error in it propagates directly into Ktrans. This is the dominant error source, and motivates both AutoAIF and the independent verification afforded by AIFArtist
- A native T1 map is required. An erroneous T1 yields an erroneous concentration and hence an erroneous Ktrans, without indication of failure
- Temporal resolution must be adequate. Models that resolve flow require high temporal resolution to resolve the bolus peak. Models without this term are more tolerant low low temporal resolution.
- Model misspecification, if the assumptions of the model are violated (e.g. no backflux in the Patlak model) the results will have large errors.
- Absolute values reported in the literature vary between institutions more than theory predicts, largely in consequence of the preceding factors.
Selection¶
| Semi-quantitative | Pharmacokinetic | |
|---|---|---|
| Arterial input function required | No | Yes |
| Native T1 map required | No | Yes |
| Units | Arbitrary | Physical |
| Comparable between institutions | No | In principle |
| Sensitivity to temporal resolution | Low | High |
| Resolves perfusion from permeability | No | Yes |
| Failure mode | Apparent | Frequently silent |
The governing consideration is the scope of the intended comparison. Multi-center studies, longitudinal assessment of treatment response, and any analysis in which a measurement is compared against one acquired elsewhere require parameters expressed in physical units. Where comparison is confined to a single examination, or between lesions in one subject on one scanner, semi-quantitative measures are robust and their failure modes are readily apparent.
The two approaches are not mutually exclusive
A quantitative pipeline yields the semi-quantitative measures without additional acquisition, the concentration curves being already computed and IAUC being an integral of them. Reporting both is advisable. Where a Ktrans map appears anomalous, the semi-quantitative maps provide the most direct means of distinguishing a genuine physiological finding from a fitting failure, since they do not depend upon the input function or the T1 map that would otherwise be implicated.
Module E of ROCKETSHIP supports this comparison, and each fit is accompanied by a goodness-of-fit map.