Non-invasive ICP estimators

1. Bedside vignettes: why this matters in the PICU#

Vignette A. The 10-year-old in a peripheral ED#

A 10-year-old presents to a rural ED 90 minutes from the nearest pediatric neurosurgical centre after a bicycle helmet-strike. GCS 11, headache, vomiting, photophobia. The CT shows a small extradural with effacement of the basal cisterns. There is no neurosurgeon, no invasive monitor, and a 90-minute transfer ahead. You scan the left MCA with a 2 MHz probe: PSV 95, EDV 18, PI 1.7. ONSD on the right is 6.2 mm, on the left 6.0 mm. The Bellner regression (ICP ≈ 10.93 × PI − 1.28) returns ICP ≈ 17 mmHg, but the ONSD is significantly above the pediatric cut-off. You treat empirically with 3% saline 3 mL/kg, head up 30°, and accelerate the transfer with a paramedic crew briefed on the working diagnosis. 1 2

Vignette B. The 2-month-old with bulging fontanelle on the post-natal ward#

A 2-month-old infant is seen on the post-natal ward 6 weeks after discharge with vomiting, poor feeding, and a tense fontanelle. The closest centre with a pediatric neurosurgeon is 4 hours away. There is no CT in the unit. Through-fontanelle ultrasound shows ventriculomegaly. ONSD (still measurable through the open suture but better imaged from the transorbital window): 5.5 mm bilateral, well above the < 1 y cut-off. TCD through the fontanelle: PI 1.8 in the MCA. The combination triggers an immediate phone call to the pediatric neurosurgical service, transfer arranged with hypertonic saline ready, and the infant is in theatre 6 hours later for an EVD. 2 3

Vignette C. The adolescent with idiopathic intracranial hypertension and "normal" PI#

A 16-year-old with new-onset chronic headaches, papilloedema on fundoscopy, and a normal CT presents for evaluation of IIH. ONSD is 6.3 mm bilaterally. TCD PI is 0.9 (normal). The team is briefly reassured by the PI but the optometry exam, the symptoms, and the ONSD are concordant for raised ICP. PI in chronic, low-grade raised ICP is often normal because the cerebrovascular bed has time to adapt; ONSD is the more sensitive index here. Lumbar puncture confirms opening pressure 32 cmH2O. 4 5

2. What NI-ICP is, and what it is not#

Invasive ICP measurement (parenchymal Codman / Camino strain gauge probes, intraventricular EVD) is the standard against which all non-invasive methods are validated. The non-invasive methods are estimators: they use a proxy signal (Doppler velocity, optic sheath diameter, tympanic-membrane displacement, skull pulsation amplitude) and convert it via a regression or a waveform-shape algorithm to an estimated ICP.

Three things follow.

Accuracy is method-, age-, and centre-dependent. Most NI-ICP estimators have 95% confidence intervals of ±10 to 15 mmHg around the true ICP. They are sufficient to exclude severe intracranial hypertension with a high NPV (Rasulo's 2022 IMPRESSIT-2 multicentre study: the TCD-derived ICP estimate excludes intracranial hypertension with a high negative predictive value, though with poor quantitative concordance, ~33%) but they are not sufficient to measure ICP in the way an invasive monitor does. They do not let you target a precise CPP. 6 3

Trend over absolute. A rising PI from 0.9 to 1.5 within a single child within an hour is a useful signal even when the absolute number is not diagnostic. An ONSD that has grown from 5.0 to 6.0 mm over 12 hours in the same child is more informative than the absolute value. Document with the same probe, same operator where possible.

Multimodal beats single-modality. TCD PI and ONSD together perform better than either alone: PI catches the acute haemodynamic signature of raised ICP; ONSD catches the structural (CSF-pressure) signature. Add fontanelle ultrasound in the infant. Add clinical exam always. 7 4

3. The methods: anatomy and principle#

3.1 TCD pulsatility index (PI)#

PI = (PSV − EDV) / MFV. Rises when distal cerebrovascular resistance rises, including from raised ICP. The Bellner 2004 regression in 81 neurosurgical patients:

with 95% CI ±10 to 12 mmHg around the predicted value. 1 8 The TCD-derived ICP estimate has a high NPV for excluding severe intracranial hypertension but poor quantitative concordance with invasive ICP (~33%), so it rules raised ICP out rather than measuring it; the PPV for diagnosing raised ICP is poor (~0.50). 6 9 In children specifically, a PI threshold of 1 detected only ~25% of cases with ICP > 20 mmHg (sensitivity ~25%, specificity ~88%), so PI is unreliable as a pediatric ICP estimate. 10

Strengths: bedside, fast, available wherever there is a TCD probe. Limitations: PI rises with anything that increases distal resistance (hypocapnia, hyperoxia, hypothermia, low arterial compliance in neonates), not only ICP.

3.2 Optic nerve sheath diameter (ONSD)#

The optic nerve is surrounded by a subarachnoid sheath continuous with the intracranial subarachnoid space. CSF pressure transmitted along this sheath distends it. Measurement: 7 to 12 MHz linear probe over a closed eyelid; identify the optic nerve as a hypoechoic stripe posterior to the globe; measure the sheath outer-to-outer diameter at exactly 3 mm posterior to the retinal surface. Both eyes; take the average of two readings per side. 11 12

Age-banded cut-offs:

Strengths: bedside, repeatable, inexpensive, validated against invasive ICP with sensitivity ~85% and specificity ~75% at the appropriate age-banded threshold; a pediatric meta-analysis reports higher pooled accuracy (sensitivity ~92%, specificity ~89%). 14 Limitations: operator-dependent (intra-rater variability up to ±0.3 mm); ethnic and sex variation; not real-time (lag of minutes-to-hours behind acute ICP changes); falsely elevated in optic neuritis, Graves' orbitopathy, and post-orbital trauma.

3.3 Brain4Care extensometer#

A scalp-applied strain gauge that measures skull pulsation amplitude with each cardiac cycle. The recorded waveform has the same P1-P2-P3 morphology as the invasive ICP waveform. The ratio P2/P1 rises with falling intracranial compliance (the same physiology that drives the invasive ICP-waveform abnormality). The device returns a non-invasive ICP waveform "shape" rather than an absolute number, and a derived P2/P1 ratio. Validation studies are ongoing. 15 16

Strengths: continuous, true waveform analysis, no skull penetration. Limitations: still emerging clinical evidence; sensitive to scalp contact and probe position; the absolute pressure derivation is less robust than the waveform shape.

3.4 Tympanic membrane displacement (TMD)#

The perilymph in the cochlea is continuous with the CSF via the cochlear aqueduct, so perilymph pressure tracks ICP. The stapedial reflex causes a small TM displacement whose amplitude varies with perilymph pressure. Method: sealed ear-probe, acoustic stimulus elicits stapedial reflex, the device measures TM displacement to ~0.1 nm resolution. Promising in posture-change ICP changes; less established for absolute ICP. 3 Strengths: continuous, painless. Limitations: requires intact TM and middle ear; cochlear aqueduct closure with age (~50% of adults have closed aqueducts) limits applicability.

3.5 Two-depth TCD (Vittamed)#

A specialised TCD that measures simultaneously at two depths (intracranial and extracranial portions of the ophthalmic artery), inferring the pressure gradient across the orbit and hence ICP. Requires a calibrated external pressure cuff over the orbit. 17 3 Strengths: directly calibrated, less regression-dependent than PI. Limitations: device cost, operator training, limited availability.

3.6 MRI / CT-based volumetric estimators#

Quantitative CSF flow MRI (phase-contrast at the aqueduct), ventricular volume change with posture (MRI), CT-based volumetric techniques. Research tools mostly; not bedside.

3.7 Other methods#

• Otoacoustic emissions (OAEs) for ICP-related cochlear pressure changes (research).
• Pupillometry indirectly: a falling NPi is consistent with raised ICP but is not a measurement; the NPi quantifies the pupillary light reflex for trend-following. 18
• Fundoscopy / papilloedema: late, chronic sign of raised ICP; useful for the IIH and chronic-ICP context but not acute. Papilloedema is present in only ~20% of children with raised ICP, so its absence does not exclude raised ICP. 19

4. The signal: what to record at the bedside#

Document the operator (intra-rater consistency matters), the probe (for repeat ONSD), and the patient state (head position, sedation, PaCO2).

5. What is normal? Age-banded reference values#

1 2 4 11.

6. What is abnormal? Pattern library and method comparison#

Decision tree: "should I act?"#

flowchart TD
  Clin[Clinical suspicion of raised ICP] --> Avail{Methods available?}
  Avail -->|TCD only| PIcheck[PI]
  Avail -->|ONSD only| ONSDcheck[ONSD by age]
  Avail -->|Both| Both[PI + ONSD]
  PIcheck -->|PI low| Low[Raised ICP unlikely]
  PIcheck -->|PI raised| Higher[Higher probability; combine with exam]
  ONSDcheck -->|Below age cut-off| Norm[Reassure]
  ONSDcheck -->|Above age cut-off| Abn[Abnormal; treat or escalate]
  Both -->|Both abnormal| Treat[Empirical treatment; transfer or invasive monitor]
  Both -->|One abnormal| Recheck[Recheck both in 1 h; consider escalation]
  Both -->|Both normal| Trend[Continue serial monitoring]

7. Try it: interactive widgets#

8. Management decisions driven by NI-ICP#

NI-ICP does not titrate CPP. It triggers escalation decisions: empirical treatment, invasive monitor placement, neurosurgical transfer.

8.1 The triage ladder#

1. Both estimators normal + reassuring exam: continue observation, repeat at planned interval.
2. One estimator abnormal: empirical head-of-bed, normocapnia, normothermia; recheck both in 1 h.
3. Both estimators abnormal OR rapid trend change: empirical hypertonic saline or mannitol; arrange CT or repeat imaging; consider invasive monitor placement; involve neurosurgery.
4. Both estimators strongly abnormal + clinical deterioration: full empirical treatment (osmotherapy, head up, sedation, normocapnia); transfer for invasive monitoring and definitive care.

8.2 Empirical osmotherapy thresholds#

A child with PI > 1.4, ONSD above the age cut-off, and a GCS that has fallen by 2 points warrants empirical hypertonic saline 3% at 3 mL/kg over 15 minutes. The pediatric BTF guidance accepts empirical osmotherapy when invasive monitoring is delayed and clinical signs concur. 20

8.3 Transfer decisions#

In a peripheral ED, a child with concordant NI-ICP abnormalities and concerning imaging is a high-priority transfer to a neurosurgical centre. The NI-ICP estimators help in the transfer briefing: "PI 1.7 bilateral, ONSD 6.5 mm bilateral, treated with 3% saline; ICP probably 25 to 35 in our estimation; ETA your unit 60 minutes."

8.4 Centre comparison#

In centres without invasive monitoring capability, NI-ICP estimators (especially the TCD-ONSD pairing) are the de facto monitoring standard. The pediatric MMM consensus places them as a tier-2 modality in resource-stratified settings. 21 22

9. Clinical contexts: NI-ICP across acute brain injuries#

9.1 Severe TBI in resource-limited or pre-transfer settings#

The most common use case. Pre-hospital and emergency-department NI-ICP (TCD PI + ONSD) identifies children whose intracranial pressure is likely elevated enough to merit empirical treatment and rapid neurosurgical referral. Tazarourte's 2011 cohort and Bouzat's 2014 study showed pre-hospital TCD PI predicts in-hospital ICU outcome. 23 24 6

9.2 Aneurysmal SAH#

Post-SAH hydrocephalus and vasospasm both raise ICP. NI-ICP estimators can monitor between formal CT scans, especially in children where serial CT carries cumulative radiation risk. PI rises with vasospasm independently of ICP, complicating interpretation; ONSD is the cleaner ICP signal in this context. 25 26

9.3 Pediatric AIS with malignant oedema#

A child with a large MCA infarct can develop malignant oedema 2 to 5 days post-stroke. NI-ICP estimators triage the decision for decompressive hemicraniectomy: rising PI, expanding ONSD, and exam deterioration together justify operative intervention even without a pre-existing ICP probe. 27

9.4 HIE and post-cardiac arrest#

Diffuse cerebral oedema in severe HIE raises ICP, sometimes profoundly. NI-ICP estimators inform decisions about head positioning, mild hyperventilation as a temporising measure, and consideration of EVD placement in selected cases. ONSD is easier than TCD in the chubby neonatal scalp; fontanelle ultrasound adds a third dimension in infants. 28 29

9.5 Pediatric ECMO#

Daily ONSD on VA-ECMO is part of some units' neurological surveillance bundles. A rising ONSD over 24 hours in a comatose ECMO patient triggers head CT or fontanelle ultrasound (in infants). The non-pulsatile circulation of full-flow ECMO makes TCD PI less useful (PI falls toward zero when pulsatility is lost). 30 31

9.6 Meningitis and encephalitis#

A child with bacterial meningitis and a tense fontanelle (infant) or evolving GCS depression (older child) warrants ONSD and TCD as part of the empirical raised-ICP workup. Cerebral oedema in fulminant meningitis can necessitate EVD placement; the NI-ICP signals support the decision when imaging cannot be obtained quickly. 32 33

9.7 Brain-death determination#

NI-ICP estimators are not part of formal brain-death protocols. TCD waveform analysis (oscillating / pendular / systolic-spike-only patterns) is a recognised ancillary test in some jurisdictions. ONSD is not used for brain death. 34 35

9.8 DKA cerebral oedema#

In a child being rehydrated for DKA, ONSD measurements during the high-risk window (hours 4 to 24 of treatment) can detect early oedema. Rising ONSD beyond the age cut-off in a symptomatic child warrants immediate osmotherapy. PI is less reliable here because hyperglycaemia and rehydration alter cerebral haemodynamics. 36 37

9.9 Idiopathic intracranial hypertension (IIH)#

ONSD is the NI-ICP of choice in chronic IIH because TCD PI is often normal (cerebrovascular adaptation). Serial ONSD complements fundoscopy and LP opening pressure in the long-term monitoring of IIH treatment response. 3 4

10. Multimodal integration: NI-ICP in the MMM/MNM stack#

11. Setup and technique#

11.1 TCD PI measurement#

1. Probe: 2 MHz pulsed-wave (blind TCD) or 2 to 4 MHz TCCD with B-mode.
2. Position: temporal window, depth 30 to 50 mm depending on age, MCA M1 segment.
3. Optimise: angle < 30°, loud crisp spectral envelope, gain set so the envelope is bright without saturation.
4. Sample: average PSV, EDV, MFV over ≥ 5 cardiac cycles per side.
5. Compute: PI = (PSV − EDV) / MFV.
6. Apply Bellner regression: ICP ≈ 10.93 × PI − 1.28 (95% CI ±10 to 12 mmHg).
7. Document trend: serial measurements every 1 to 4 hours.

11.2 ONSD measurement#

1. Probe: 7 to 12 MHz linear ultrasound probe (the same probe used for vascular access).
2. Position: patient supine, gentle eye-closure, generous gel cushion to minimise globe pressure.
3. Image: hypoechoic optic nerve in long axis behind the globe.
4. Measure: outer-sheath to outer-sheath diameter exactly 3 mm posterior to the retinal surface.
5. Average: two measurements per side, average across sides if symmetric, or report each side.
6. Apply age-banded cut-off: < 1 y ~4.0 mm; 1 to 15 y ~4.5 mm; adult ~5.0 to 5.7 mm.
7. Be gentle: do not press on the globe; pseudo-elevation from probe pressure is a real artefact.

11.3 B4C setup#

1. Skin preparation: clean a small area of parietal scalp; the device contacts the skull directly.
2. Device placement: secure with the manufacturer's headband; check signal quality on the device screen.
3. Calibrate: per device protocol (may include zero reference to baseline period).
4. Record: continuous waveform; the device reports P1, P2, P3 amplitudes and the P2/P1 ratio.
5. Trend: monitor P2/P1 trajectory over hours; rising P2/P1 ratio = falling compliance, even when absolute pressure is unclear.

11.4 TMD setup#

1. Insert sealed ear-probe in the external auditory canal, ensure tight seal.
2. Acoustic stimulus triggers stapedial reflex.
3. Measure TM displacement amplitude.
4. Interpret per device protocol (research / specialty centres mainly).

11.5 Two-depth TCD (Vittamed)#

1. Position the device at the orbital window with the calibrated external pressure cuff.
2. Inflate cuff in steps; the device measures velocity in the intracranial and extracranial ophthalmic artery segments simultaneously.
3. The cuff pressure at which the two velocity envelopes match is the ICP estimate (the "pressure balance point").
4. Report with confidence interval; requires trained operator.

11.6 Quality control#

• Operator training is the single biggest determinant of NI-ICP accuracy. Maintain a small group of trained operators per unit.
• Local validation against any invasive ICP placements in your centre over time refines local cut-offs.
• Inter-observer agreement studies are part of any NI-ICP programme; intra-class correlation > 0.8 is the goal.

12. Pitfalls#

• PI is not ICP. PI rises with hypocapnia, hyperoxia, hypothermia, vasoconstrictors, neonatal low arterial compliance, and many other things. Always interpret in context.
• ONSD probe pressure artefact. Pressing the probe on the globe transiently raises measured sheath diameter; use generous gel and minimal pressure.
• ONSD inter-rater variability up to ±0.3 mm; significant when the action threshold is 4.5 mm in a child.
• B4C absolute pressure is less reliable than the waveform shape; trend P2/P1, not the inferred mmHg.
• TMD requires intact TM and patent cochlear aqueduct; ~50% of adults have closed aqueducts.
• Two-depth TCD requires expensive specialised equipment and a trained operator.
• Chronic raised ICP (IIH) has normal PI because of vascular adaptation; ONSD is the better signal here.
• VA-ECMO removes pulsatility; PI falls toward zero regardless of ICP.
• In an infant, the open fontanelle vents ICP partially; ONSD remains useful but interpret with the fontanelle ultrasound context.
• Single-estimator triage: never make a treatment decision on one estimator alone; always combine.
• Sedation effect on PI: deep sedation lowers MFV (CMRO2 effect) but does not change PI much; light sedation can produce sympathetic surges that raise PI without raising ICP.

13. Combine with…#

• TCD: the parent modality for PI; same probe, same windows.
• ONSD: the dedicated ONSD measurement page.
• ICP: the invasive gold standard against which NI-ICP is validated.
• Pupillometry: early-warning brainstem index alongside PI / ONSD.
• Fontanelle ultrasound: the infant adjunct.
• Foundations: Monro-Kellie doctrine: the physiology behind every NI-ICP signal.

14. Evidence summary#

15. Recent literature (2022 to 2025)#

• Brasil 2022: TCD waveform analysis for non-invasive ICP via Brain4Care-style extensometers; transition from research to bedside. 16
• Cardim 2023: multicentre validation of non-invasive ICP estimators against parenchymal probes. 39
• Rasulo 2022 (IMPRESSIT-2): multicentre study showing TCD reliably excludes intracranial hypertension (high NPV) but with poor quantitative concordance with invasive ICP. 6 Rasulo 2024: Brain4Care P2/P1 ratio correlates with invasive ICP in adult TBI; pediatric data emerging. 15
• Figaji 2025 pediatric MMM consensus: positions NI-ICP estimators as tier-2 modalities in resource-stratified pediatric centres. 21
• Robotic / continuous TCD platforms: enable continuous PI monitoring without operator presence; relevant for trend-based NI-ICP triage.
• ONSD pediatric updates: post-2020 pediatric data continue to support the 4.5 mm threshold in 1 to 15 y children, with some centres advocating ethnicity-adjusted cut-offs.

16. Self-check#