Chapter 6 — Advanced Modalities: Strain, 3D, Contrast and Artificial Intelligence

📚 Guideline basis

EACVI/ASE/Industry Task Force to standardise deformation imaging; ASE/EACVI recommendations for image acquisition and display using three-dimensional echocardiography; Porter TR, et al. 2018 ASE guidelines update on ultrasound enhancing agents (JASE 2018;31:241–74); Narang A, et al. on artificial intelligence in echocardiography (JASE 2023, PMID 37059617); ASE 2025 diastolic and right heart guidelines for strain thresholds.

Speckle-tracking strain

Physics

Ultrasound speckle — the interference pattern generated by sub-resolution scatterers within myocardium — is stable enough between consecutive frames to be tracked. Displacement of speckle kernels over the cardiac cycle yields deformation.

ε (Lagrangian) = ((L - L₀) / L₀) × 100%

Longitudinal strain is negative because the myocardium shortens; more negative is better. Global longitudinal strain (GLS) averages segmental peak systolic longitudinal strain across the three apical views.

Acquisition requirements

Normal values and the vendor problem

Normal GLS is approximately −20% or more negative. RV free wall strain and RV global strain are normal when more negative than approximately −20% to −25%, with the ASE 2025 right heart guideline grading dysfunction from those values (Chapter 9). LA reservoir strain ≤ 18% indicates elevated LA pressure in the ASE 2025 diastolic algorithm (Chapter 8).

Strain values are vendor-dependent. Absolute values differ between analysis packages by amounts comparable to the difference between normal and mildly abnormal. The ASE 2025 diastolic guideline reports LA strain normal ranges separately by vendor for this reason. Two practical rules follow:

1. Never compare a strain value obtained on one vendor's package with a threshold derived on another.
2. For serial monitoring, use the same machine and the same software version, and say so in the report.

Strain in critical illness

⚠️ Load dependence

GLS is load-dependent in the same direction as ejection fraction: it improves with vasodilatation and worsens with vasoconstriction. Serial GLS across a large change in noradrenaline dose is not comparing like with like, and this is not addressed by any current guideline.

Three-dimensional echocardiography

Practical constraints in the ICU: multibeat "stitched" acquisitions produce stitch artefacts with arrhythmia or ventilation-induced translation. Single-beat acquisitions avoid this at the cost of volume rate. Target volume rates above 20 Hz for structural assessment, and prefer single-beat acquisition when measurements will be taken.

Ultrasound enhancing agents

Covered operationally in Chapter 3. The essentials:

• Indication: ≥ 2 contiguous non-diagnostic LV segments; suspected LV thrombus; apical variant HCM; apical aneurysm; incomplete TR envelope for PASP estimation (agitated saline for the latter).
• Technique: low mechanical index (0.1–0.3). High MI destroys microbubbles.
• Artefacts: swirling apical defect from excessive MI mimicking thrombus; basal attenuation from excessive concentration.
• Under-use is the problem, not over-use. Contrast frequently converts a non-diagnostic ICU study into a diagnostic one and avoids escalation to TEE.

Artificial intelligence

What exists

Why ICU images are the hard case

Every failure mode of these systems is concentrated in exactly the population this book addresses.

1. Training-data distribution. Models are overwhelmingly trained on ambulatory laboratory studies: sinus rhythm, good windows, standard views, resting loading. ICU images are off-axis, foreshortened, low-frame-rate, arrhythmic and acquired through hyperinflated lungs. Performance degrades, and the degradation is not always signalled by a confidence score.
2. Silent failure. A segmentation model given a foreshortened apical view returns a plausible EF. It does not report that the view was foreshortened. The error is invisible in the output.
3. Physiological blindness. An automated EF of 65% in vasoplegic septic shock is numerically correct and clinically misleading (Chapter 29). No current system reasons about loading conditions.
4. Guideline inheritance. Automated diastolic grading built on the 2016 algorithm reproduces its indeterminate rate and its exclusion violations — it will happily grade a patient in atrial fibrillation.

The governance rule

Every automated measurement used in a clinical decision must be verified by the operator against the image from which it was derived. In practice: check the contour, check the view is not foreshortened, check the timing, and check the number against the visual impression. Where they disagree, the image wins.

Report automated measurements as such, and state the software used. This is a reproducibility requirement, and it also makes downstream audit possible.

⚠️ Evidence quality

No AI echocardiography system has been shown in a randomised trial to improve patient outcomes in critical care. The evidence base is diagnostic-accuracy studies, largely in non-ICU populations. Regulatory clearance establishes acceptable performance against a reference standard in a defined population; it does not establish benefit in yours.

Where AI plausibly helps in the ICU

• Reducing measurement variability for serial monitoring, where consistency matters more than absolute accuracy — automated VTI tracing on the same patient across a fluid challenge is a good fit.
• Guidance for less experienced operators acquiring standard views out of hours.
• Quality assurance triage, flagging studies for expert over-read (Chapter 42).

ICU-specific limitations — summary

🛑 Critical pitfall: Comparing serial strain values acquired on different machines or software versions. The vendor difference can exceed the physiological change being sought.

🛑 Critical pitfall: Accepting an automated ejection fraction without inspecting the contour and confirming the apical view is not foreshortened. Foreshortening inflates EF by 5–10 points and the algorithm will not tell you.

🛑 Critical pitfall: Reporting a 3D volume from a stitched multibeat acquisition in a patient with atrial fibrillation. The stitch artefact invalidates the measurement.

• 💡 Clinical pearl: 3D LVOT planimetry is the most under-used advanced technique in critical care haemodynamics. It removes the squared-diameter error that dominates every stroke volume calculation in this book.
• 💡 Clinical pearl: RV free wall strain detects dysfunction before TAPSE falls and is not abolished by pericardiotomy — making it the RV measure of choice after cardiac surgery.
• 💡 Clinical pearl: Treat AI output as a second opinion from a colleague who cannot see the patient. Useful, worth checking, never final.

References

1. Voigt JU, Pedrizzetti G, Lysyansky P, et al. Definitions for a common standard for 2D speckle tracking echocardiography: consensus document of the EACVI/ASE/Industry Task Force. Eur Heart J Cardiovasc Imaging 2015;16:1–11.
2. Lang RM, Badano LP, Tsang W, et al. EAE/ASE recommendations for image acquisition and display using three-dimensional echocardiography. J Am Soc Echocardiogr 2012;25:3–46.
3. Porter TR, Mulvagh SL, Abdelmoneim SS, et al. Clinical applications of ultrasonic enhancing agents in echocardiography: 2018 ASE guidelines update. J Am Soc Echocardiogr 2018;31:241–74.
4. Narang A, et al. Artificial intelligence in echocardiography. J Am Soc Echocardiogr 2023. PMID 37059617.
5. Nagueh SF, Sanborn DY, Oh JK, et al. ASE 2025 diastolic function update. J Am Soc Echocardiogr 2025;38:537–69.
6. American Society of Echocardiography. Guidelines for the echocardiographic assessment of the right heart in adults. J Am Soc Echocardiogr 2025.