DSC Perfusion MRI — rCBV Normalisation ROI Placement

MRIninja Knowledge Base | Focus / Deep Dive Page Version 1.0 — April 2026 Companion pages: DSC MR Perfusion Parts I–III; DSC Clinical Indications and Interpretation

1. The Question and Why It Matters

When calculating normalised relative cerebral blood volume (nrCBV), the raw rCBV value of the tumour is divided by the rCBV of a reference region in normal-appearing white matter (NAWM). This step converts the arbitrary post-processing units of the raw rCBV map into a clinically interpretable ratio — the value that drives the diagnostic thresholds used throughout neuro-oncology practice (nrCBV ≥ 1.75 for high-grade transformation; < 1.0 for treatment effect).

The choice of where to place this reference ROI has direct consequences for the numerical result. A poorly placed reference region produces a systematically shifted nrCBV value that can move a tumour from "probable progression" to "indeterminate" — or vice versa — based purely on measurement methodology rather than biology. This is not a minor technical detail: it is one of the most common sources of inter-reader and inter-institution variability in clinical DSC perfusion, and it is a problem that the literature has addressed directly in the past decade.

The specific question asked here — should the reference ROI always be in the contralateral centrum semiovale, or should it follow the tumour to lower anatomical levels? — has a well-supported answer, with nuances that depend on tumour location.

2. Why Normalisation Is Necessary

Raw rCBV values from DSC perfusion are in arbitrary scanner-dependent units. They vary between:

• Scanner manufacturers and field strengths
• TE and TR choices
• Leakage correction methods
• Post-processing software platforms

Without normalisation, an rCBV value of 3.8 at one institution cannot be compared to 3.8 at another. By dividing the tumour rCBV by a reference tissue rCBV from the same scan, most of these sources of variability cancel out, because they affect the numerator and denominator equally [1].

The reference region must therefore be:

• In normal-appearing white matter (no tumour infiltration, no radiation-related changes, no prior treatment effect)
• In the same hemisphere or contralateral hemisphere (not in territory potentially affected by the lesion)
• Reproducibly identifiable across readers and across serial time points
• In an area with negligible susceptibility artefact on the EPI acquisition

3. Regional Variation in White Matter rCBV: The Biological Foundation of the Question

White matter does not have a uniform rCBV throughout the brain. This is the central biological fact that makes the question clinically important.

Normal cerebral blood volume varies systematically by:

Anatomical level: Supratentorial white matter has higher rCBV than infratentorial white matter. Within the supratentorial compartment, deep white matter (centrum semiovale, corona radiata) has different blood volume from periventricular white matter and subcortical U-fibres.

Proximity to grey matter: White matter adjacent to cortex (subcortical U-fibres, subgyral white matter) has higher rCBV than deep white matter, because the highly vascular cortex and its penetrating vessels contribute partial volume signal to the adjacent white matter on DSC maps.

Age: Normal white matter rCBV declines approximately 3–6% per decade, meaning reference values are age-dependent [2]. An older patient's NAWM rCBV is systematically lower than a younger patient's, which would — if not accounted for — produce artificially elevated nrCBV in older patients using a population-level reference.

Posterior fossa: Cerebellar white matter and brainstem have different rCBV from supratentorial white matter. The posterior fossa is also the region most affected by susceptibility artefacts on GRE-EPI, making reference ROI placement in the posterior fossa particularly unreliable. For sequence-level protocol optimisation, vendor terminology and artefact management, see the dedicated MRIninja page Gradient Echo (GRE/FLASH) Sequence.

Temporal lobe: The temporal lobe white matter has higher rCBV than the centrum semiovale, partly due to partial volume averaging with the overlying highly vascular temporal cortex and amygdala. A reference ROI placed in the contralateral temporal white matter at the level of a temporal lobe tumour would systematically overestimate the NAWM reference value, producing artificially low nrCBV.

This regional variability is why the intuitive approach — "place the ROI in the white matter at the same level as the tumour, contralaterally" — is technically flawed for caudally located tumours.

4. The Accumulated Evidence: What Studies Have Shown

4.1 Centrum Semiovale as Superior Reference: Reproducibility Data

The most directly relevant data come from the 2018 European Radiology study by Smits et al. [3], which compared multiple reference tissue locations (contralateral NAWM by free choice, contralateral NAWM at the tumour hotspot level, centrum semiovale, putamen, thalamus, and several others) across multiple readers. The key finding: centrum semiovale was the only reference tissue that showed excellent intra- and inter-observer agreement (ICC > 0.85) and lowest coefficients of variation (< 12.5%) across all tested locations.

Contralateral NAWM at the tumour level, by comparison, showed substantially lower reproducibility — particularly when the tumour was located in the temporal lobe, where available contralateral white matter at the same anatomical level is limited and partially contaminated by partial-volume averaging with grey matter structures.

The 2024 study by Coban et al. [4], evaluating inter-reader reproducibility across neuroradiologists, general radiologists, and residents, confirmed this finding quantitatively: in an analysis of all strategies, the ICC for the centrum semiovale was significantly higher than that for contralateral white matter (P < 0.001).

The 2022 multi-reader study by Leu et al. [5] directly compared four normalisation methods across 35 glioma patients — single-plane CSO, multi-sphere CSO, single-plane at tumour level, multi-sphere at tumour level. All four normalisation methods had similar intra-reader repeatability and inter-reader reproducibility, but there were significant reductions in time when performing the CSO methods compared to the tumour-level methods. One likely explanation for the increased time to create NAWM ROIs in tumour regions was for cases where the tumour was located in regions with minimal contralateral white matter, such as near subcortical structures or in the temporal lobes.

The study concluded that the present findings may support the usage of the centrum semiovale as a target NAWM region instead of the white matter directly opposite the tumour, because the centrum semiovale is a large, homogeneous, consistently available reference area that is minimally affected by partial-volume contamination with grey matter or susceptibility artefact.

4.2 The Multicentre Reproducibility Literature

The NCI Quantitative Imaging Network (QIN) multisite concordance studies are the highest-quality evidence base for DSC reproducibility. The Schmainda et al. 2018 study [6] and the accompanying QIN digital reference object study [7] both demonstrate that NAWM reference ROI placement is one of the dominant sources of inter-site rCBV variability when site-specific software and ROI placement are used. Standardised rCBV (srCBV), which replaces manual NAWM normalisation with a histogram-based training set approach, was developed specifically to remove the dependence on ROI placement [8]. This approach entirely bypasses the question being discussed here — but requires validated training data and compatible software, and is not universally available clinically.

4.3 The Serial Imaging Argument

For serial studies assessing treatment response — which is the most clinically demanding normalisation context — consistency of the reference ROI location across timepoints is as important as the absolute location choice. The NAWM reference region must be placed in exactly the same anatomical location at every follow-up examination. Any shift in reference ROI position between scans introduces artefactual apparent nrCBV change that mimics or masks true treatment effect.

The centrum semiovale satisfies this serial consistency requirement better than tumour-level white matter for two reasons: it is anatomically stable and easily identifiable across all supratentorial cases, and it is not in the radiation field for the large majority of supratentorial gliomas — protecting it from radiation-induced perfusion changes that would confound the reference value over time.

5. The Practical Answer: A Location-Stratified Framework

There is no single universally correct answer. The optimal reference ROI strategy depends on tumour location, and each scenario has a specific recommendation supported by the evidence.

5.1 Supratentorial Tumours (Frontal, Parietal, Occipital, Central)

Recommendation: Contralateral centrum semiovale, posterior third preferred.

This is the standard and best-validated reference location for all supratentorial tumours not in the temporal lobe. The posterior centrum semiovale (at the level of the body of the corpus callosum, posterior to the central sulcus) is preferred because:

• It is distant from frontal and temporal lobe grey matter that increases partial-volume contamination in the anterior centrum semiovale
• It is away from the periventricular region where CSF pulsation artefacts may occasionally affect rCBV
• It is in a consistent anatomical position that can be reproducibly identified across readers and visits

Multiple ROIs (3–5) placed anteriorly to posteriorly in the centrum semiovale — averaged — provide better reproducibility than a single ROI and reduce the impact of any local signal inhomogeneity [4, 5].

5.2 Temporal Lobe Tumours

Recommendation: Contralateral centrum semiovale (NOT the contralateral temporal white matter).

Temporal lobe white matter has significantly higher rCBV than centrum semiovale white matter, due to partial-volume averaging with amygdala, hippocampus, and temporal cortex. If the reference ROI is placed in the contralateral temporal white matter at the tumour level, it will overestimate NAWM rCBV, producing artificially low nrCBV values for the tumour.

For temporal lobe tumours, placing the reference ROI in the contralateral centrum semiovale — even though it is at a more cranial level than the tumour — produces more reproducible and biologically meaningful nrCBV values. This is confirmed by the Leu et al. 2022 study, where temporal lobe location was specifically identified as the scenario where tumour-level NAWM ROI placement is most difficult and most error-prone [5].

Practical note: when a temporal lobe tumour extends inferiorly near the skull base, the DSC slice coverage at that level may itself be compromised by susceptibility artefacts from the petrous bone and mastoid. The reference ROI must be verified to be in a region with intact signal on the source EPI images — not in a distorted or dropped-out zone.

5.3 Deep/Subcortical Tumours (Basal Ganglia, Thalamus, Internal Capsule)

Recommendation: Contralateral centrum semiovale — with additional caution about grey matter contamination.

Deep tumours near the basal ganglia, thalamus, and internal capsule present two reference placement challenges:

• The "contralateral equivalent location" contains grey matter (contralateral thalamus, putamen) with inherently high rCBV, which would dramatically inflate the denominator and produce falsely low nrCBV
• The surrounding white matter at this level (posterior limb of the internal capsule, posterior periventricular white matter) is thin and difficult to sample without grey matter partial-volume contamination

The correct reference for deep tumours remains the centrum semiovale. The posterior limb of the internal capsule has been used as an alternative in some published series but shows higher variability than centrum semiovale in reproducibility studies [3].

5.4 Posterior Fossa Tumours (Cerebellum, Brainstem)

Recommendation: Contralateral centrum semiovale — do not use posterior fossa reference.

Posterior fossa white matter should never be used as the reference for DSC normalisation. Two specific problems make it unsuitable:

Susceptibility artefact: GRE-EPI sequences show severe signal dropout and geometric distortion adjacent to the petrous bone, mastoid air cells, and posterior fossa skull base. The source EPI images must be inspected before accepting any perfusion measurement from the posterior fossa — the reference region must be confirmed to be in undistorted signal.

Different perfusion physiology: Cerebellar white matter and brainstem have different autoregulatory properties and baseline rCBV from supratentorial white matter. Using infratentorial white matter as reference for a supratentorial tumour would not be meaningful, and using it as reference for a posterior fossa tumour introduces a non-validated normalisation that has no established threshold data.

For posterior fossa tumours, the supratentorial centrum semiovale is used as reference — the same as for all other locations. The nrCBV thresholds established in glioma literature are calibrated to centrum semiovale normalisation, and using a different reference invalidates their application.

The limitation is acknowledged: a tumour at the level of the pons with a reference in the centrum semiovale means the reference and tumour are at different Z-levels. This is unavoidable and acceptable, because the alternative — a posterior fossa reference with poor reproducibility and susceptibility artefacts — is worse.

5.5 Summary Decision Table

6. Practical ROI Placement Rules

Regardless of tumour location, the following rules apply universally to reference ROI placement in clinical DSC perfusion:

• Always verify the source EPI image at the reference ROI level before accepting the reference rCBV value. The region must show uniform, undistorted signal. If there is susceptibility dropout, the rCBV in that area is unreliable.
• Avoid cortex: the reference ROI must be entirely within white matter, not overlapping the grey-white junction. On DSC maps (low resolution, 2–3 mm voxels), apparent "white matter" can easily include partial volume grey matter signal. Place the ROI visually in the centre of the white matter, not at its margin.
• Avoid CSF spaces: periventricular white matter immediately adjacent to the lateral ventricles can include partial-volume CSF signal, which is T2-bright and would reduce the measured rCBV, inflating the tumour/reference ratio.
• Avoid vessels: large pial arteries visible on the rCBV map as intensely bright focal spots must not be included in the reference ROI. These produce artefactually elevated local rCBV from saturation effects.
• ROI size: 40–60 mm² is the standard range used in published reproducibility studies [3, 4]. Smaller ROIs increase sampling variance; larger ROIs increase the risk of grey matter inclusion.
• Multiple ROIs: using 3–5 ROIs placed across the centrum semiovale (anterior, middle, posterior) and averaging them significantly improves inter-reader reproducibility compared to a single ROI [4]. This is the recommended approach for serial studies and clinical trials.
• Consistency on serial studies: on every follow-up scan, the reference ROI must be placed in the same anatomical location as the baseline examination. Any change in reference location between timepoints introduces artefactual apparent nrCBV change.

7. When the Standard Approach Is Insufficient: Standardised rCBV

Manual NAWM normalisation — including optimised centrum semiovale placement — remains subject to operator-dependent variability because the exact ROI boundaries and location cannot be perfectly reproduced between readers or between institutions. The inter-reader reproducibility of rCBV using even the best manual normalisation approach remains only moderate (ICC 0.51–0.53 between experienced neuroradiologists [4]), which is a genuine limitation for clinical decision thresholds.

Standardised rCBV (srCBV), introduced by Bedekar, Jensen and Schmainda in 2010 [8], addresses this by replacing the manual NAWM ROI with a histogram normalisation approach trained on a reference dataset. The entire rCBV map is transformed so that the distribution of rCBV values in normal brain matches a population reference standard. This eliminates the need for a manual reference ROI entirely, removing operator variability from the normalisation step. The Schmainda et al. 2019 QIN validation [1] demonstrated that srCBV and nrCBV achieve statistically equivalent performance for tumour rCBV classification when BSW leakage correction is applied.

srCBV requires access to validated training data and compatible post-processing software (IB Neuro is the most widely used clinical implementation). In centres with this capability, srCBV offers superior reproducibility for multicentre and longitudinal applications. In centres without it, rigorously applied centrum semiovale nrCBV is the appropriate standard.

8. Implications for Threshold Interpretation

The diagnostic thresholds for nrCBV (1.75 for high-grade transformation; 1.0 as the lower boundary of indeterminate range) were established in studies that used centrum semiovale normalisation [9, 10]. If a department uses tumour-level contralateral white matter as the reference for temporal or posterior fossa tumours, the resulting nrCBV values are not comparable to these published thresholds.

This is a critical point: using non-standard reference regions does not just affect reproducibility — it shifts the numerical values in a predictable direction and invalidates direct comparison with published threshold data. A temporal lobe glioma with nrCBV of 1.8 using centrum semiovale normalisation and nrCBV of 1.4 using ipsilateral temporal white matter normalisation represent the same biological reality, but one meets the published threshold for probable high-grade transformation and the other falls in the indeterminate zone.

Departments that use non-standard reference locations must either:

• Establish their own threshold values through institutional validation against histopathology, or
• Switch to centrum semiovale normalisation to align with published threshold data

9. Radiation Field Considerations for Serial Studies

In post-treatment glioma follow-up, the white matter within the radiation field undergoes progressive changes over time — including radiation-induced leukoencephalopathy, gliosis, and perfusion alterations — that may affect the rCBV of the reference white matter if it is within the high-dose radiation volume.

For this reason, the reference ROI must be placed in white matter outside the radiation field. The centrum semiovale of the contralateral hemisphere is outside the radiation field for the majority of unilateral supratentorial gliomas. Confirming this on the radiation therapy plan is advisable in ambiguous cases.

If the contralateral centrum semiovale itself is within the radiation field (bilateral treatment, whole-brain radiation, extended field radiation), the reference ROI must be documented carefully and its potential contamination acknowledged in the report.

12. Evidence-Based References

A. Guidelines / Consensus / Society Recommendations

B. Systematic Reviews / Meta-analyses

No systematic reviews specifically addressing reference ROI placement were identified. The evidence base is primarily from prospective reproducibility studies.

C. Important Prospective / Original Studies

D. Technical MRI Papers

E. Landmark Historical References

End of document — MRIninja v1.0 — April 2026

Companion pages: DSC MR Perfusion Parts I–III; DSC Clinical Indications and Image Interpretation