Multimodal Echocardiographic Assessment in Intravenous Immunoglobulin‐Resistant Kawasaki Disease: From Prediction to Long‐Term Surveillance

ABSTRACT

Kawasaki disease (KD) with intravenous immunoglobulin (IVIG) non‐response is a high‐risk state in which fever response, coronary evolution, and myocardial dysfunction may diverge. This narrative review, appraised against SANRA after drafting, examines echocardiography across the disease‐management lifecycle, from pre‐treatment risk enrichment to long‐term surveillance. PubMed/MEDLINE searches were supplemented by cross‐source discovery and metadata verification. Standardized serial coronary measurement with consistent use of one Z‐score system remains the management core. Initial composite echocardiographic abnormalities, myocardial deformation, and tissue‐Doppler indices may enrich risk assessment, but current discrimination is moderate and thresholds lack external validation. IVIG non‐response and coronary aneurysm are related but non‐interchangeable; defervescence does not establish coronary safety, whereas definite baseline coronary artery aneurysm is a treatment‐relevant high‐risk phenotype. Deformation, myocardial work, right‐heart indices, coronary flow reserve, and complementary imaging can refine selected clinical questions but should not trigger treatment in isolation. Echocardiography can therefore structure coronary and myocardial surveillance, but cannot yet independently determine treatment resistance or support a universal multimodal treatment score. Prospective multicenter standardization, external validation, and clinical‐utility testing are required before multimodal imaging‐guided treatment decisions are adopted.

A structured lifecycle echocardiographic framework for intravenous immunoglobulin (IVIG)‐resistant or high‐risk Kawasaki disease spans pre‐treatment coronary phenotyping to long‐term surveillance, integrating serial coronary and myocardial assessment to guide anatomy‐adapted management. Serial two‐dimensional imaging with Z‐scores is the cornerstone; defervescence alone does not establish coronary safety.

1. Introduction

Kawasaki disease (KD) is an acute childhood vasculitis in which coronary artery dilation, aneurysm, thrombosis, stenosis, and myocardial involvement create a continuum of short‐ and long‐term cardiovascular risk. Contemporary American Heart Association guidance recognizes both the centrality of cardiac imaging and the remaining uncertainty regarding the optimal frequency and modality of serial assessment in patients at risk for progressive coronary disease or inducible ischemia [1]. The 2017 scientific statement established the modern Z‐score‐based coronary risk framework, while the 2024 update incorporated subsequent evidence on acute imaging, high‐risk phenotypes, and long‐term management [1, 2]. Japanese cardiovascular‐sequelae guidance and the American College of Rheumatology/Vasculitis Foundation guideline add complementary perspectives on risk‐tiered follow‐up and treatment of high‐risk or refractory disease [3, 4].

Approximately 10%–20% of treated children remain or become febrile after initial IVIG in commonly cited clinical series, although definitions and event rates vary across settings [5, 6]. This group is clinically important because treatment resistance and early cardiovascular abnormalities may coexist, yet the prediction problem remains unresolved. Risk scores developed in Japan have shown limited sensitivity in geographically and ethnically distinct populations, and adding a statistically significant marker does not necessarily produce a calibrated, transportable, or clinically useful prediction model [6, 7, 8]. A key practical question is therefore whether echocardiography can contribute more than the detection of an already established aneurysm.

Beyond the limited transportability of Japanese scores, non‐response to initial IVIG appears to track the intensity of the acute systemic inflammatory phenotype and may coexist with coronary involvement, although these trajectories are not deterministic. In Turkish and Italian cohorts, higher admission inflammatory burden—including C‐reactive protein—was associated with IVIG non‐response, while coronary abnormalities followed partly distinct risk patterns. These observations support the biological rationale for risk enrichment outside Asian populations, but they do not establish universally transportable thresholds [9, 10].

Routine echocardiography has traditionally centered on coronary lumen dimensions and conventional ventricular function. That foundation remains indispensable, but newer methods can interrogate myocardial deformation, pressure–strain work, right‐heart performance, coronary flow physiology, and temporal changes in cardiovascular phenotype. These measurements have generated a growing but uneven evidence base. Some studies directly compare IVIG non‐responders with responders, whereas many others address coronary artery lesions (CAL), coronary artery aneurysms (CAA), myocardial dysfunction, or late vascular abnormalities. Treating these outcomes as interchangeable would overstate both predictive accuracy and clinical actionability.

Previous statements and modality‐oriented reviews have established how echocardiography is performed and how coronary involvement is classified [1, 2, 11]. The present review takes a different organizing approach: it follows the management lifecycle and asks what an imaging result can legitimately change at each stage. Direct evidence for IVIG non‐response is therefore kept separate from evidence for CAA, myocardial phenotype, or late ischemia, and every proposed decision node is labelled according to evidentiary maturity.

This narrative review therefore examines echocardiography as a lifecycle decision‐support platform in children with IVIG‐resistant KD or imaging‐defined high‐risk phenotypes. It covers conventional two‐dimensional and Doppler assessment, coronary Z scores, speckle‐tracking‐derived strain, myocardial work, right ventricular (RV) function and RV–pulmonary arterial (PA) coupling, and transthoracic coronary flow reserve (CFR). Cardiovascular magnetic resonance (CMR) and intravascular ultrasound (IVUS) are discussed only where they validate, expose limitations of, or prompt escalation beyond echocardiography. Rather than recapitulating general KD diagnosis or pharmacological management, the review addresses three aims: to synthesize echocardiographic markers of IVIG non‐response and high‐risk cardiovascular phenotypes; to compare the maturity and complementary roles of multimodal assessment across the disease lifecycle; and to propose an evidence‐calibrated, explicitly unvalidated framework for acute risk stratification, treatment monitoring, and long‐term surveillance.

2. Review Scope and Search Strategy

This review was prepared as a transparent narrative synthesis. The Scale for the Assessment of Narrative Review Articles (SANRA) was applied after drafting as a six‐domain completeness appraisal covering justification of importance, aims, search description, referencing, scientific reasoning, and presentation of endpoint data [12]. SANRA did not serve as a prospective protocol, reporting checklist, study‐level risk‐of‐bias instrument, or evidence‐grading system; eligibility and interpretation were governed by the review question and the outcome‐specific criteria described below.

PubMed/MEDLINE was searched without a date restriction through July 1, 2026. Core concepts included “Kawasaki disease,” “intravenous immunoglobulin resistance,” “IVIG non‐response,” “initial echocardiogram,” “coronary Z score,” “global longitudinal strain,” “speckle tracking,” “myocardial work,” “right ventricular,” “coronary flow reserve,” “intravascular ultrasound,” “cardiac magnetic resonance,” “pulse‐wave velocity,” and “carotid intima–media thickness.” PubMed discovery was supplemented by cross‐source searching in OpenAlex, Europe PMC, and Semantic Scholar. Targeted update searches covered guidelines, pediatric measurement standards, treatment‐response discordance, long‐term coronary events, stress echocardiography, artificial intelligence, and radiomics. Backward and forward citation chasing was used for landmark and validation studies. PMID and DOI identifiers and bibliographic metadata were verified against PubMed and Crossref.

Priority was given to contemporary scientific statements, systematic reviews, multicenter or externally validated studies, studies directly comparing IVIG non‐responders with responders, and imaging studies reporting sample size or diagnostic performance. Reports without an imaging outcome, isolated case reports, and non‐peer‐reviewed preprints were not used as core evidence. No formal language restriction was applied to database queries; synthesis required sufficient English‐language metadata, abstract, or full text for reliable interpretation. Because searches were theme based and intentionally overlapping, module‐specific yields were not summed into a PRISMA‐style total. This was not a systematic review, and no claim is made that every eligible report was retrieved. Evidence directly addressing IVIG non‐response was separated from evidence addressing CAL, CAA, myocardial phenotype, or long‐term vascular outcomes. Retained search terms, reproducible queries, and theme‐specific approximate yields are provided in Table S1 and the accompanying search‐strategy supplement. Table 1 presents key study characteristics including IVIG non‐response definitions, time windows, treatment era, and initial regimen to facilitate transparent appraisal of between‐study heterogeneity and contemporary applicability.

TABLE 1.

Echocardiographic studies relevant to early identification of IVIG non‐response or a high‐risk coronary phenotype in Kawasaki disease.

A. direct evidence for IVIG non‐response
Study	Design and population	Treatment era / initial regimen	Echo timing / index	IVIG non‐response definition / time window	Main result	Validation and principal limitation
Chbeir et al. [7]	Retrospective cohort at one French center; n = 157; 45 non‐responders	2018 publication; initial 2 g/kg IVIG reported. Recruitment period and rescue regimen not specified in this table source.	First echo at median fever Day 7; composite of mild coronary dilation, PVB, CAA, effusion, LV dysfunction, or mitral regurgitation	Fever >38.0°C at 48 h after IVIG; 6‐week CAL also assessed.	Full cohort: abnormal echo 22/45 (49%) versus 26/112 (23%); sensitivity 49%, specificity 77%. Pretreatment subset n = 58: 12/16 (75%) versus 17/42 (40%); sensitivity 75%, specificity 60%	Only 58 children had pretreatment echo; univariable testing only; no AUC, adjusted effect, calibration, or validation; composite effect cannot be assigned to individual components
Bar‐Meir et al. [5]	Nine‐center development n = 312; independent five‐center validation n = 186; additional small prospective cohort, Israel	2018 publication; recruitment years and initial regimen not reported in the accessible abstract.	Initial coronary abnormality or presentation before fever Day 5; exact echo timing and coronary definition not reported in accessible abstract	IVIG non‐response; exact fever threshold, time window, and event counts not reported in the accessible abstract.	AUC 0.70 (95% CI 0.60–0.80); sensitivity 81% (95% CI 67%–90%); specificity 50% (95% CI 44%–56%)	External validation and prospective testing reported, but dataset‐specific performance, coefficients, calibration, missing‐data handling, and net benefit were not available from the abstract
Wang et al. [30]	Matched comparison at one Chinese center: resistant n = 50, responsive n = 50, controls n = 50	2019 publication; recruitment years and initial regimen not reported in the source reviewed for this table.	Pretreatment 2D‐STE; absolute GLS <16.8%, absolute GCS <15.9%	Persistent/recurrent fever ≥38.3°C from >36 h through Day 7 after IVIG.	GLS: AUC 0.78, sensitivity 83%, specificity 69%; GCS: AUC 0.75, sensitivity 79%, specificity 65%. Interobserver ICC: 0.962 for GLS, 0.786 for GCS	Thresholds selected and evaluated in the same fixed 1:1 sample; no multivariable adjustment, calibration, resampling, external validation, or AUC confidence intervals
Phadke et al. [30]	Age‐matched comparison at one US center: resistant n = 51, responsive n = 51	Children's Hospital Colorado, February 2007–March 2014; initial regimen not reported in the source reviewed for this table.	Diagnostic TDI; tricuspid, septal, and mitral annular e′	Study‐defined IVIG resistance; exact fever threshold and time window not reported in the source reviewed for this table.	AUC 0.66 for tricuspid e′, 0.66 for septal e′, and 0.70 for mitral e′	Fixed matched sampling; thresholds evaluated in the same cohort; no calibration, incremental‐value analysis, or external validation
McCandless et al. [32]	Pretreatment comparison: acute KD n = 32, including 14 resistant cases; controls n = 22	2013 publication; study period and initial regimen not reported in the source reviewed for this table.	Velocity‐vector‐derived global longitudinal strain and strain rate	Persistent/recrudescent fever after treatment; exact threshold/time window not reported; coronary dilation also analyzed.	Resistant subgroup had lower longitudinal strain and strain rate than controls despite normal routine systolic indices	Small subgroup analysis; association rather than a prediction model; no threshold, AUC, adjustment, or validation
Lam et al. [6]	Korean cohort n = 7101 and San Diego cohort n = 649	2024 publication; cohort‐specific treatment era and initial regimen were not reported in the source reviewed for this table.	Routine clinical/laboratory features with or without worst coronary Z score on initial echo or during hospitalization	IVIG resistance; exact fever threshold and time window not reported in the source reviewed for this table.	Maximum median AUC 0.711 and 0.696; adding coronary Z score did not materially improve either cohort	Echo timing could extend beyond a purely pretreatment landmark; cross‐validation within cohorts rather than a portable echo‐based rule; useful negative incremental‐value evidence

B. Indirect evidence for an imaging‐defined high‐risk phenotype
Study	Design and population	Treatment era / initial regimen	Index	Target outcome	Main result	Correct interpretation
Jari and Esmaeili [33]	Cross‐sectional; n = 58	2024 publication; study period and initial regimen not reported in the source reviewed for this table.	Ocular inflammation and coronary dilation	Coronary dilation and IVIG resistance	Uveitis associated with both outcomes	Multisystem inflammatory context; not a prediction‐model study
Son et al. [8]	North American cohort; n = 261	2017 publication; study period and initial regimen not reported in the source reviewed for this table.	Baseline zMax ≥2.0	Subsequent CAA	CAA 16% versus 2%; C statistic 0.77; sensitivity 80%; NPV 98%	Predicts CAA, not IVIG non‐response; prevalence‐dependent NPV cannot be transported uncritically
Wang et al. [34]	Data‐driven cluster analysis; n = 1016	2023 publication; study period and initial regimen not reported in the source reviewed for this table.	Clinical, laboratory, treatment‐day, and baseline coronary variables	Clinical subgroups with IVIG‐resistance and CAA patterns	Highest‐resistance subgroup had lowest CAA rate; highest‐CAA subgroup had lowest resistance rate	Demonstrates phenotype heterogeneity and endpoint non‐equivalence; not an individual prediction rule
Matsuoka et al. [46]	Cohort restricted to clinical IVIG responders; n = 325	2020 publication; study period and initial regimen not reported in the source reviewed for this table.	Baseline zMax and 4‐week coronary trajectory	Persistent or de novo CAA at 4 weeks	13/325 had CAA; zMax >2.5 was the sole independent risk factor in single‐dose responders (OR 9.08, sensitivity 50%, specificity 91%)	Wide confidence interval; restricted responder cohort; supports continued coronary surveillance, not prediction of IVIG non‐response

Abbreviations: AUC, area under the receiver‐operating‐characteristic curve; CAA, coronary artery aneurysm; CAL, coronary artery lesion; GCS, global circumferential strain; GLS, global longitudinal strain; ICC, intraclass correlation coefficient; IVIG, intravenous immunoglobulin; LV, left ventricular; NPV, negative predictive value; OR, odds ratio; PVB, perivascular brightness; STE, speckle‐tracking echocardiography; TDI, tissue Doppler imaging.

For transparent evidence synthesis, we used prespecified narrative evidence‐status labels rather than formal GRADE ratings. Guideline‐supported core denotes a modality or parameter explicitly recommended for routine baseline or serial risk assessment by at least one major KD clinical‐practice guideline. Cohort‐supported adjunct denotes replicated or multicenter observational evidence for a clinically relevant phenotype, but no externally validated treatment threshold or demonstrated clinical utility. Exploratory and exploratory adjunct denote predominantly small, single‐center, technical‐feasibility, or non‐externally validated evidence; adjunct indicates selective use rather than routine decision‐making. Selective physiological or long‐term adjunct denotes a test reserved for a defined flow‐limitation, ischemia, or long‐term‐surveillance question, whereas complementary/reference and mechanistic/reference modalities clarify questions beyond routine echocardiography. Selective/research and proof‐of‐concept denote preliminary evidence insufficient for universal surveillance or clinical deployment.

3. A Layered Technical Foundation for Lifecycle Assessment

3.1. The Foundational Layer: Standardized Anatomy and Conventional Function

The foundational examination comprises reproducible two‐dimensional visualization of the left main coronary artery, left anterior descending artery, and right coronary artery; body‐size‐adjusted coronary dimensions; left ventricular (LV) systolic function; regional wall motion; valvular regurgitation; and pericardial findings. Pediatric quantification recommendations provide the general acquisition and measurement framework [13]. Serial assessment is more informative than isolated classification because both the maximum and current coronary dimensions influence management [1, 2, 3, 11]. Conventional schedules at diagnosis, 1–2 weeks, and 4–6 weeks provide general KD anchors, but they should not be presented as a schedule specifically validated for IVIG‐resistant cohorts [11]. Progressive abnormalities or high‐risk anatomy require more individualized imaging intensity under contemporary guidance [1, 3].

Coronary Z scores are not measurement‐neutral. Reference systems differ in population, body‐size modelling, and treatment of non‐linearity. The Dallaire–Dahdah equations and the Japanese lambda–mu–sigma approach are influential examples of distinct modelling strategies [14, 15], while the Pediatric Heart Network normal echocardiogram database illustrates the broader importance of age, sex, race, and ethnicity when pediatric Z scores are interpreted [16]. Comparative KD studies show that formula selection can materially alter the classification of dilation and aneurysm [17, 18]. Measurement and formula variability can also reclassify a substantial proportion of patients, particularly near management boundaries [19]. The clinically defensible approach is therefore to record acquisition quality, raw dimensions, body‐surface‐area equation, and Z‐score formula; use the same system longitudinally for an individual patient; and interpret borderline changes against measurement uncertainty and trajectory. Switching formulas during follow‐up can create a false biological signal.

3.2. The Functional Enhancement Layer

The functional layer includes two‐dimensional speckle‐tracking echocardiography (2D‐STE), layer‐specific strain, myocardial work, diastolic indices, RV function, and RV–PA coupling. These measures can identify abnormalities when LV ejection fraction is preserved, but their biological meaning and technical maturity differ. A 2026 meta‐analysis of 34 studies supports selective abnormalities in GLS, GCS, mitral annular e′, and left atrial reservoir strain, while overall differences in E/e′, global radial strain, and left atrial volume index were not significant [20]. Strain is influenced by loading conditions, age, frame rate, vendor, and analysis software; myocardial work reduces but does not eliminate load dependence because it incorporates estimated pressure rather than invasively measured LV pressure. RV indices are particularly sensitive to image quality and geometry. Accordingly, these parameters should be viewed as phenotype refiners rather than interchangeable components of a single score.

3.3. Problem‐Directed Physiological and Cross‐Modality Escalation

Transthoracic Doppler CFR can add physiological information about epicardial stenosis or microvascular dysfunction [21, 22], while CMR can characterize ventricular mechanics, scar, perfusion, and ischemia [23, 24, 25]. Cardiac computed tomographic angiography (CTA) can clarify distal coronary anatomy, thrombus, and stenosis that are incompletely visualized on echocardiography. In 18 patients with 21 paired examinations, proximal absolute measurements agreed reasonably well, but discrepancies were greater distally and CTA led to a lower AHA risk category in 24% of studies [26]. IVUS and virtual‐histology IVUS can expose vessel‐wall remodeling that is invisible to lumen‐only measurements [27, 28, 29]. These methods are not equivalent extensions of routine echocardiography. CFR requires specialized acquisition and has been evaluated in small, selected cohorts; CTA involves ionizing radiation and its dimensions should not be substituted uncritically into echocardiographic Z‐score systems; CMR may be limited by availability, scan duration, and the need for sedation in younger children; IVUS is invasive and should be reserved for a catheterization‐based clinical question. Their value lies in resolving a specific uncertainty, not in being applied uniformly to all high‐risk children. Table 2 summarizes each modality's target, evidence status, proposed role, and main limitation. Figure 1 illustrates these complementary layers and their current evidentiary maturity.

TABLE 2.

Multimodal cardiovascular imaging and vascular phenotyping in Kawasaki disease: Targets, evidence maturity, limitations, and proposed roles.

Modality/parameter	Target	Representative evidence	Evidence status	Proposed role	Main limitation
2D coronary measurement and Z scores	Coronary lumen size and trajectory	[1, 2, 3, 8, 9, 10, 11, 13, 14, 15, 17, 18, 19]	Guideline‐supported core	Baseline and serial risk classification	Formula‐ and measurement‐dependent reclassification
2D‐STE GLS/GCS	Global myocardial mechanics	[20, 30, 31, 41]	Cohort‐supported adjunct	Acute high‐risk myocardial phenotyping and serial functional assessment	No validated treatment threshold; vendor/load dependence
Layer‐specific strain	Transmural myocardial mechanics	[39, 40]	Exploratory	Detect subclinical and residual myocardial dysfunction	No harmonized pediatric thresholds
Myocardial work	Pressure–strain‐derived LV work	[38, 43]	Exploratory adjunct	Selected cases with suspected dysfunction despite preserved LVEF/GLS	Small studies; mainly CAL outcome; estimated pressure
RV function and RV–PA coupling	Right‐heart performance and afterload matching	[45, 46]	Exploratory	Selected high‐risk or hemodynamically abnormal patients	No external validation or hard‐outcome evidence
Transthoracic CFR	Epicardial/microvascular flow physiology	[21, 22]	Selective physiological adjunct	Specialist evaluation of suspected flow limitation	Small selected cohorts; spectrum bias; technical expertise
Stress echocardiography	Inducible wall‐motion abnormality and ischemic risk	[54]	Selective long‐term adjunct	Patients with important coronary sequelae or suspected ischemia	Small selected cohort; older testing era; threshold not broadly validated
Three‐dimensional echocardiography	Ventricular synchrony; potential future coronary geometry	[37]	Exploratory	Research	Direct KD evidence does not validate 3D coronary aneurysm morphology
IVUS/VH‐IVUS	Coronary wall architecture and remodeling	[27, 28, 29]	Mechanistic/reference	Catheterization‐based clarification of wall or lesion pathology	Invasive; not a routine echo escalation
CMR	Scar, perfusion, ventricular mechanics, complex anatomy	[23, 24, 25]	Complementary/reference	Problem‐directed advanced assessment	Availability, scan duration, sedation, selection bias
Cardiac CTA	Distal coronary anatomy, thrombus, and stenosis	[26]	Complementary/reference	When echocardiographic coronary visualization is incomplete	Radiation; modality‐dependent dimensions; no interchangeable CT Z‐score system
PWV/cIMT/FMD	Long‐term systemic vascular phenotype	[55, 56, 57]	Selective/research	Long‐term research or individualized vascular assessment	High heterogeneity; not a universal clinical endpoint
AI‐based coronary analysis	Automated classification or brightness quantification	[58, 59]	Proof of concept	Research on reproducibility and image‐derived phenotypes	Small one‐center datasets; low specificity or unsupervised endpoints; no external clinical‐utility validation

Notes: Evidence‐status definitions: These are prespecified narrative labels, not formal GRADE ratings. Guideline‐supported core requires explicit endorsement for routine baseline or serial risk assessment by at least one major KD clinical‐practice guideline. Cohort‐supported adjunct requires replicated or multicenter observational evidence for a clinically relevant phenotype, but no externally validated treatment threshold or demonstrated clinical utility. Exploratory and exploratory adjunct denote predominantly small, single‐center, technical‐feasibility, or non‐externally validated evidence; adjunct indicates selective rather than routine use. Selective physiological/long‐term adjunct denotes use for a defined flow‐limitation, ischemia, or long‐term‐surveillance question. Complementary/reference and mechanistic/reference clarify questions beyond routine echocardiography. Selective/research and proof‐of‐concept denote evidence insufficient for universal surveillance or clinical deployment.

Abbreviations: AI, artificial intelligence; CFR, coronary flow reserve; cIMT, carotid intima–media thickness; CMR, cardiovascular magnetic resonance; CTA, computed tomographic angiography; FMD, flow‐mediated dilation; IVUS, intravascular ultrasound; PA, pulmonary arterial; PWV, pulse‐wave velocity; RV, right ventricular; VH‐IVUS, virtual‐histology intravascular ultrasound.

FIGURE 1.

Complementary imaging layers and evidentiary maturity in coronary and myocardial assessment. The foundational layer comprises standardized two‐dimensional coronary imaging, Doppler assessment, and a consistently applied Z‐score system. The functional adjunct layer includes GLS/GCS, layer‐specific strain, myocardial work, diastolic indices, and RV–PA coupling. The advanced or complementary layer includes transthoracic CFR, stress echocardiography, CMR, CTA, and IVUS/VH‐IVUS. IVUS is displayed as an invasive mechanistic reference rather than a routine echocardiographic escalation step; CMR and CTA are cross‐modality tools for scar, ischemia, complex or distal anatomy, thrombus, and stenosis. Three‐dimensional coronary aneurysm reconstruction and AI/radiomics are classified as exploratory. Layer position reflects clinical maturity and intended role, not diagnostic superiority.

4. Evidence Across the Disease‐Management Lifecycle

4.1. Acute Phase: Predicting IVIG Non‐Response

The most accessible signal is not a single subjective coronary sign but an abnormal composite examination. In a retrospective cohort of 157 children, 31% had an abnormal first echocardiogram at a median of 7 days of fever. The composite included coronary dilation or aneurysm, pericardial effusion, perivascular brightness, LV dysfunction, or mitral regurgitation. IVIG non‐response, defined as temperature above 38.0°C at 48 h after infusion, occurred in 45 children. A composite abnormality was present in 49% of non‐responders versus 23% of responders (p = 0.002), corresponding to sensitivity of 49% and specificity of 77% [7]. The temporal sequence limits its interpretation as a prediction study: among 117 children with complete timing data, only 58 underwent echocardiography before IVIG, whereas the remainder were examined after treatment. In that pretreatment subgroup, sensitivity was 75% and specificity was 60%. The study therefore supports exploratory risk enrichment, not an independently validated pretreatment rule, and it cannot assign the composite effect to individual components.

A multicenter study developed and externally evaluated a simple model based on presentation before Day 5 or an initial coronary abnormality. Reported discrimination for IVIG non‐response was moderate (area under the curve [AUC] 0.70, 95% confidence interval 0.60–0.80), with sensitivity of 81% and specificity of 50% [5]. Such a profile may support early risk enrichment, particularly when the cost of missing a high‐risk child is considered greater than the burden of closer observation. It is insufficient for imaging‐only diagnosis of treatment failure or automatic escalation of anti‐inflammatory therapy.

2D‐STE provides the most direct functional evidence identified in the present search. A comparative study at one center included 50 IVIG‐resistant children, 50 responders, and 50 healthy controls; echocardiography was performed before IVIG. Non‐response was defined as persistent or recurrent temperature of at least 38.3°C from more than 36 h through Day 7 after the infusion. An absolute global longitudinal strain (GLS) below 16.8% yielded an AUC of 0.78, sensitivity of 83%, and specificity of 69%; an absolute global circumferential strain (GCS) below 15.9% yielded an AUC of 0.75, sensitivity of 79%, and specificity of 65% [30]. Reproducibility was assessed in 20 examinations: interobserver intraclass correlation coefficients were 0.962 for GLS and 0.786 for GCS. These estimates suggest that myocardial mechanics capture information related to the resistant inflammatory phenotype. However, the fixed 1:1 sampling does not reproduce clinical prevalence, thresholds were selected and evaluated in the same cohort, and no multivariable adjustment, calibration, resampling validation, or external validation was reported.

Two earlier deformation and Doppler studies point in the same direction but do not resolve validation. A velocity‐vector‐imaging study of 32 children with acute KD, including 14 IVIG‐resistant patients, associated the resistant phenotype with more abnormal myocardial strain and strain rate, but it was designed to detect group‐level myocardial abnormalities rather than produce a transportable prediction rule [31]. In an age‐matched comparison of 51 non‐responders and 51 responders, lower tricuspid, septal, and mitral annular e′ velocities yielded AUCs of 0.66, 0.66, and 0.70, respectively [32]. The balanced sampling, modest discrimination, and lack of external validation again favor adjunctive phenotyping over threshold‐based treatment.

A much larger negative study is particularly informative. Machine‐learning models using 7101 Korean cases and 649 San Diego cases achieved maximum median AUCs of 0.711 and 0.696, respectively. Adding the worst coronary Z score from the initial echocardiogram or hospitalization did not materially improve models based on routine laboratory data [6]. Thus, an echocardiographic feature can be clinically important for coronary management without adding enough independent information to predict fever‐defined treatment resistance.

Indirect evidence should be labelled as such. A cross‐sectional study of 58 children linked ocular inflammation with coronary dilation and IVIG resistance, supporting a possible multisystem inflammatory phenotype but not an echocardiographic prediction model [33]. Baseline maximum coronary Z score (zMax) of at least 2.0 was associated with subsequent CAA, with a reported C statistic of 0.77, sensitivity of 80%, and negative predictive value of 98% [8]. This is important risk‐stratification evidence, but the target outcome was later CAA rather than IVIG non‐response. Data‐driven clustering of 1016 children makes the distinction concrete: one subgroup had the highest IVIG‐resistance rate but the lowest CAA rate, whereas a younger subgroup had the highest CAA rate but the lowest IVIG‐resistance rate [34]. Resistance and coronary injury overlap, but neither can serve as a surrogate for the other.

Perivascular brightness (PVB) illustrates why biological plausibility must be separated from measurement reliability. Earlier work did not establish PVB as a reliable stand‐alone diagnostic sign [35]. In a blinded multi‐reader study, inter‐reader reliability was only 0.13, and the sign could also be observed in controls with other febrile illnesses [36]. PVB may remain part of an overall abnormal examination, but it should not be used as an isolated predictor or a mandatory decision node. Similarly, the current evidence does not support a reusable, independently validated E/e′ threshold for IVIG non‐response.

Taken together, acute echocardiography can augment clinical risk assessment but cannot yet replace it. The most defensible approach is to use standardized coronary findings and composite cardiovascular abnormalities to identify patients who warrant closer clinical and serial imaging assessment, while treating strain as an adjunct in experienced centers. A universal “echocardiography plus Kobayashi/Egami” score has not been externally validated across populations.

4.2. Acute and Subacute Coronary Phenotyping

Once high risk is suspected, the clinical question shifts from prediction to phenotype: what coronary abnormality is present, how quickly is it evolving, and what complication must be excluded? Current guidance keeps standardized coronary dimensions and serial Z‐score trajectories at the center of this assessment [1, 2, 3]. Baseline zMax adds prognostic information for subsequent CAA, but its interpretation depends on a consistent formula and should not be conflated with an IVIG‐response endpoint [8]. Formula‐comparison and measurement‐variability studies reinforce that apparent prevalence, risk category, and even management classification can change when methods are altered [17, 18, 19]. Each report should therefore preserve the raw dimensions, segment, image quality, body‐surface‐area equation, Z‐score system, maximum historical Z score, and current Z score.

CFR moves beyond lumen diameter to functional coronary assessment. In an early transthoracic study, a CFR threshold of 2.2 or less identified left anterior descending coronary stenosis with reported sensitivity of 90% and specificity of 100% [21]. These striking values arose from a small, selected sample and are vulnerable to spectrum effects. A separate study of 25 children suggested late coronary microvascular dysfunction even without overt CAL [22]. CFR is therefore best positioned as a specialist or research adjunct when symptoms, anatomy, or other testing raises a question of flow limitation; it is not a general screening threshold for IVIG resistance.

IVUS studies provide a mechanistic explanation for why apparent luminal regression is not equivalent to vascular normalization. Conventional IVUS demonstrated persistent intimal thickening and abnormal wall morphology after KD [27]. Virtual‐histology studies in adolescents and young adults and in lesions assessed early and more than a decade after KD showed complex remodeling and tissue characteristics that may persist despite improvement in lumen dimensions [28, 29]. These observations strengthen the rationale for long‐term risk‐tiered surveillance, but IVUS is an invasive reference method rather than a routine echocardiographic escalation step.

Three‐dimensional echocardiography is often presented as an intuitive route to aneurysm morphology and volume. The direct KD evidence identified here, however, mainly concerns real‐time three‐dimensional assessment of LV systolic dyssynchrony rather than validated three‐dimensional coronary aneurysm reconstruction [37]. Coronary 3D reconstruction should therefore remain a research frontier until acquisition success, reproducibility, incremental value, and clinical impact are demonstrated.

4.3. Myocardial and Right‐Heart Phenotyping

Myocardial inflammation can be clinically occult when conventional ejection fraction is preserved. The direct IVIG‐resistance evidence for abnormal myocardial deformation and annular e′ supports functional imaging as an early phenotype, while leaving its incremental predictive value unsettled [30, 31, 32]. Studies in broader acute and convalescent KD cohorts extend this observation. Layer‐specific strain and pressure–strain‐derived myocardial work can detect reduced myocardial performance despite apparently normal routine systolic function [38, 39, 40]. In a cohort of 101 acute KD cases, longitudinal strain improved immediately after IVIG and had returned to control values by 6–8 weeks in the serially examined subgroup; the change tracked inflammatory markers rather than coronary dilation [41]. Other serial speckle‐tracking and tissue Doppler observations also suggest partial or substantial recovery over time [42]. These studies document temporal phenotype, not a treatment‐specific causal effect.

The aggregate evidence is now broader than any single threshold study. A 2026 meta‐analysis included 34 studies, 1821 patients with KD, and 1157 controls. GLS, GCS, e′, and left atrial reservoir strain were abnormal overall, whereas pooled differences in E/e′, global radial strain, and left atrial volume index were not significant [20]. The result supports selective subclinical dysfunction but also argues against describing all systolic and diastolic measures as uniformly impaired. It does not establish a threshold for IVIG retreatment because most included studies addressed KD myocardial phenotype rather than treatment resistance.

These findings should be interpreted cautiously. A myocardial‐work study comparing 45 children with KD and coronary dilation, 10 with KD without dilation, and 45 controls reported lower myocardial work indices even when LVEF and GLS were normal [38]. Another small study reported an AUC of 0.847 for global work index in relation to CAL [43]. The latter is a CAL endpoint, not IVIG non‐response, and neither study validates myocardial work as a treatment trigger. Layer‐specific studies likewise lack harmonized pediatric reference values and cross‐vendor thresholds [39, 40].

The right heart is an under‐assessed component of the acute cardiovascular phenotype. In a cohort of 143 children, PA dilation occurred in 16.1% and was associated with abnormalities in RV myocardial performance and tricuspid E/e′ [44]. A separate 59‐patient study linked RV–PA coupling measures with CAL and reported an AUC of 0.715 [45]. These studies support systematic attention to RV size, function, and loading in selected high‐risk patients. They do not establish an externally validated RV threshold for IVIG non‐response or later hard cardiovascular events.

CMR offers complementary biological context. Follow‐up CMR can identify ventricular, perfusion, and coronary abnormalities that are not completely characterized by echocardiography [24]. Feature‐tracking studies in convalescent KD provide cross‐modality evidence of altered mechanics [25]. More recent work in 55 children with a history of KD found lower strain or strain‐rate measures among those with late gadolinium enhancement [23]. Echocardiographic deformation may therefore help select patients for CMR, but it cannot replace CMR for scar detection.

4.4. Monitoring Treatment Response and Early Evolution

No single echocardiographic parameter currently defines IVIG response. Clinical fever response remains central, while imaging describes the cardiovascular trajectory that accompanies or follows treatment. The most actionable measurements are the maximum and current coronary Z scores, new or progressive aneurysm, thrombus, stenosis, wall‐motion abnormality, ventricular dysfunction, valve regurgitation, and pericardial findings [1, 2, 3, 11]. A single post‐treatment scan can be misleading if interpreted without baseline timing and subsequent trajectory.

Fever response and coronary safety can diverge. Among 325 children who were clinically responsive to aspirin and IVIG, CAA was persistent or appeared de novo at 4 weeks in 13. In single‐dose responders, baseline zMax above 2.5 was the only independent risk factor for CAA at 4 weeks (odds ratio 9.08, 95% confidence interval 1.26–65.3; sensitivity 50%, specificity 91%) [46]. The wide confidence interval and low sensitivity preclude a reassurance rule, but the study shows why defervescence should not terminate coronary surveillance.

Baseline echocardiography can also identify a phenotype in which treatment intensification is considered. In a retrospective multicenter cohort of 121 children with CAA at diagnosis (Z score at least 2.5 but below 10), initial IVIG combined with corticosteroids or infliximab was independently associated with less subsequent progression of coronary size than IVIG alone [47]. This observational association is consistent with guideline attention to high‐risk disease [4], but it is vulnerable to residual confounding and does not validate a general imaging‐only escalation rule. Its legitimate implication is narrower: definite CAA on the pretreatment examination is management‐relevant information that should be integrated promptly with clinical risk and specialist treatment decisions.

Functional indices may enrich the trajectory. Improvement in strain or tissue Doppler mechanics after the acute phase is biologically reassuring, but observed change cannot automatically be attributed to a specific treatment rather than the natural temporal course [41, 42]. Conversely, persistence of vessel‐wall pathology on IVUS despite luminal improvement warns against equating regression of an aneurysm with complete recovery [27, 28, 29]. Serial echocardiography should therefore document both anatomy and function, while escalation to advanced imaging is driven by the unresolved clinical question rather than by an isolated experimental threshold.

4.5. Long‐Term Surveillance

Long‐term follow‐up should remain anchored to the maximum historical and current coronary risk category, with imaging intensity determined by residual anatomy, symptoms, thrombosis or stenosis, and evidence of ischemia [1, 2, 3, 48]. The reason is biological as well as administrative: luminal regression does not erase prior wall injury, and risk follows the worst coronary phenotype rather than the most reassuring single follow‐up image.

The natural‐history evidence spans treatment eras. In a classic cohort of 594 patients followed for 10–21 years, 146 had acute CAA; 28 later developed stenosis, 11 had myocardial infarction, and 5 died [49]. These estimates arose largely before contemporary IVIG and antithrombotic practice and should not be applied directly to children treated today. They nevertheless established the sequence from aneurysm to remodeling, stenosis, and ischemia that long‐term imaging seeks to intercept.

Treatment era is a material source of heterogeneity across the broader evidence base. IVIG dose and timing, aspirin practice, the use of early corticosteroids in high‐risk children, and availability of biologic rescue therapy varied substantially among the represented studies. Consequently, estimates of fever‐defined non‐response, coronary progression, and imaging performance should not be pooled or directly extrapolated to contemporary care without attention to the study‐specific regimen and recruitment period. Table 1 therefore reports the available treatment‐era and initial‐regimen information, and older cohorts are interpreted principally as natural‐history or mechanistic evidence.

Contemporary registries sharpen risk stratification. In the 1651‐patient International Kawasaki Disease Registry, 10‐year luminal narrowing, thrombosis, and major adverse cardiovascular complications were concentrated in patients with maximum CAA Z scores of at least 10; Z scores of at least 20, complex architecture, and the number and location of affected branches added risk information [50]. In a separate 1006‐patient angiography cohort with CAA, large aneurysm was the dominant correlate of later coronary events (hazard ratio 8.90), while IVIG resistance retained a smaller independent association (hazard ratio 2.20) [51]. These data link acute resistance with long‐term risk, but only within a coronary‐aneurysm population; they do not imply that every IVIG non‐responder faces the same lifelong hazard.

A modern US series of 60 patients with giant CAA reported major adverse cardiovascular events in 21.7% over a median 11 years. Bilateral aneurysms and maximum Z scores of at least 20 conferred greater risk. Although 26.7% of aneurysms regressed to a normal luminal diameter, continued surveillance remained necessary [52]. At the population level, the balance is more reassuring: among 1169 patients with KD and 11,690 matched comparators followed for a median of 20 years, the cumulative incidence of major adverse cardiac events after KD was 1.2%, despite a higher relative hazard; events occurred largely among patients with CAA, in whom cumulative incidence was 12.8% [53]. Long‐term counselling should therefore communicate both the low absolute event burden for most patients and the concentrated risk carried by prior CAA.

Echocardiography remains useful after anatomy has established risk, particularly for ventricular function and stress‐induced wall‐motion abnormalities. In 58 adolescent KD survivors, peak wall‐motion score index during dobutamine stress echocardiography independently predicted major adverse cardiac events over a mean 14.7 years; 15‐year event‐free survival was 25.0% with an index of at least 1.25 and 91.7% below that threshold [54]. The small, selected cohort and older testing era limit universal threshold adoption, but the study supports problem‐directed stress assessment in patients with coronary lesions. CMR, CTA, or catheter‐based assessment is appropriate when scar, perfusion, distal anatomy, thrombus, or stenosis cannot be adequately characterized by echocardiography [24, 25, 26]. IVUS observations further reinforce the concept that wall remodeling may persist even when the lumen appears improved [27, 28, 29].

Peripheral vascular measures have been proposed as early warning markers of later cardiovascular risk. A recovery‐phase study reported higher brachial–ankle pulse‐wave velocity (PWV) in children after KD but no corresponding difference in carotid intima–media thickness (cIMT) [55]. A meta‐analysis of 49 case‐control studies, including 2714 participants with KD and 2118 controls, found overall evidence of lower flow‐mediated dilation and higher arterial stiffness, PWV, or cIMT, but with substantial heterogeneity [56]. An earlier systematic review similarly emphasized conflicting results, especially in patients without CAA [57]. PWV, cIMT, and flow‐mediated dilation should therefore remain selective or research‐oriented phenotypes rather than universal surveillance endpoints for all adult KD survivors.

5. An Evidence‐Tiered Echocardiography‐Informed Framework

Figure 2 maps the modalities to the four management checkpoints, and Figure 3 converts that lifecycle into an author‐proposed decision pathway.

FIGURE 2.

Lifecycle positioning of multimodal echocardiography in IVIG‐resistant or high‐risk Kawasaki disease. Multimodal assessment is organized across acute, post‐IVIG/subacute, and long‐term phases. Standardized two‐dimensional coronary measurements, one consistently applied Z‐score system, conventional ventricular function, valve regurgitation, wall motion, and pericardial findings form the core. Speckle‐tracking‐derived strain, myocardial work, diastolic measures, and right‐heart indices are candidate adjuncts. CFR, stress imaging, CMR, and IVUS are selected according to the unresolved clinical question. Solid connectors denote guideline‐supported or comparatively established assessment; dashed connectors denote candidate adjuncts without prospective external validation. The pathway depicts surveillance roles and does not imply causal effects or a validated treatment algorithm.

FIGURE 3.

Proposed echocardiography‐informed decision framework for IVIG‐resistant or high‐risk Kawasaki disease. At diagnosis, standardized assessment of the left main, left anterior descending, and right coronary arteries using one Z‐score system is integrated with ventricular function, valvular regurgitation, pericardial findings, and wall motion. A composite abnormal examination supports intensified serial assessment but does not mandate treatment escalation. Definite CAA on the baseline examination is shown separately because it is a treatment‐relevant high‐risk phenotype requiring specialist integration with clinical guidance. At the early response checkpoint, fever status is interpreted together with coronary and myocardial trajectory; defervescence does not end surveillance. Subacute checkpoints use 1–2‐week and 4–6‐week anchors adapted to evolving anatomy. Advanced testing is selected according to an unresolved question, and long‐term follow‐up is tiered to maximum historical and current coronary risk. PWV, cIMT, FMD, and related vascular measures remain selective or research oriented. The framework is author proposed and has not undergone prospective validation. Baseline zMax ≥2.0 has been associated with subsequent CAA rather than validated as a predictor of IVIG non‐response; candidate GLS, e′, myocardial work, RV–PA coupling, and CFR thresholds are therefore not mandatory decision nodes.

The proposed framework is a decision‐support hierarchy, not a new prediction rule. It has three evidentiary levels: a guideline‐supported core, cohort‐supported selective adjuncts, and problem‐directed advanced or invasive assessment. The core is standardized measurement of the left main, left anterior descending, and right coronary arteries using one Z‐score system, accompanied by conventional LV and RV function, valve regurgitation, pericardial findings, and regional wall motion. GLS/GCS, myocardial work, diastolic indices, and focused RV/RV–PA assessment are selected according to the suspected phenotype and local expertise. CFR, stress imaging, CMR, CTA, and catheter‐based methods answer defined questions that the core examination cannot resolve.

The framework is applied at four checkpoints. The baseline or pretreatment checkpoint establishes raw coronary dimensions, zMax, ventricular and valvular status, pericardial findings, and an image‐quality benchmark. A composite abnormal examination warrants closer observation, while definite CAA at diagnosis is a distinct, treatment‐relevant high‐risk phenotype that should prompt specialist consideration of intensified therapy in conjunction with clinical guidance [4, 47]. Composite abnormalities, PVB, strain, e′, myocardial work, and RV–PA coupling do not independently mandate treatment.

The early response checkpoint occurs when clinical IVIG response is judged. Persistent or recurrent fever drives the definition of non‐response, while repeat echocardiography is indicated by clinical instability, a high‐risk baseline examination, or evolving cardiac findings. The subacute checkpoint uses the usual 1–2‐week and 4–6‐week anchors, adapted to coronary trajectory rather than treated as a fixed schedule validated for resistant disease [1, 2, 3]. Defervescence does not remove the need for this checkpoint because delayed CAA can occur among clinical responders [46]. At the long‐term checkpoint, maximum historical and current coronary involvement determine surveillance intensity and the need for ischemia testing or advanced anatomy [48, 50, 51]. These time points are an author‐proposed operationalization of existing evidence, not a prospectively validated protocol.

Across all checkpoints, candidate functional measures should be interpreted against age, loading conditions, vendor, image quality, and serial reproducibility. A change is most credible when acquisition and analysis are held constant and when it converges with anatomy, hemodynamics, biomarkers, or symptoms. When routine imaging does not resolve the clinical question, problem‐directed escalation can include CFR, stress echocardiography, CMR, CTA, or catheter‐based assessment. IVUS is appropriate only when invasive evaluation is otherwise clinically justified. PWV, cIMT, and related vascular measures remain selective or investigational.

The longer‐term research objective could be a dynamic multimodal model combining clinical variables, inflammatory markers, zMax, deformation, myocardial work, and right‐heart indices. Such a model should prespecify the target outcome and prediction time, use longitudinal methods for repeated examinations, report calibration as well as discrimination, undergo geographic and vendor‐diverse external validation, and demonstrate net clinical benefit before influencing treatment. The current evidence does not justify assigning weights to a unified score.

Before the four‐checkpoint framework can be considered for clinical implementation, it should be evaluated in a prospective multicenter registry enrolling consecutive children before initial IVIG. IVIG non‐response and early coronary progression should be prespecified as separate co‐primary outcomes, with exact fever definitions and imaging landmark times locked before enrolment. Acquisition should use a prespecified coronary measurement protocol, one body‐surface‐area equation and one Z‐score formula, with vendor and strain settings recorded and blinded core‐laboratory reading. Candidate models should report calibration intercept and slope, calibration plots, discrimination, Brier score, and decision‐curve analysis, followed by geographically independent external validation. Sample size should be based on anticipated event rates and the prespecified model degrees of freedom; a unified score should not be proposed until these requirements are met.

6. Critical Appraisal, Unresolved Problems, and Future Directions

6.1. Why Current Performance Estimates Do Not Justify an Echocardiography‐Based Treatment Score

The first problem is endpoint discipline. IVIG non‐response, CAL, CAA, myocardial dysfunction, and later ischemia are related but non‐interchangeable. The C statistic of 0.77 for baseline zMax, the AUC of 0.847 for myocardial work, and the AUC of 0.715 for RV–PA coupling cannot be compared as though they predict the same outcome [8, 43, 45]. The opposing IVIG‐resistance and CAA patterns in data‐derived clinical subgroups provide empirical evidence that the outcomes can dissociate [34]. Even within IVIG‐resistance studies, differences in fever‐based definitions, treatment timing, and sampling influence apparent performance.

The second problem is the distinction between association and prediction. p values for group differences or associations do not establish individual‐risk calibration. AUC summarizes ranking but does not show whether predicted probabilities are accurate, whether a marker adds value beyond a clinical and basic‐echocardiographic model, or whether acting on a threshold improves outcomes. The moderate AUC and low specificity reported for the simple multicenter model support risk triage rather than a stand‐alone treatment decision [5]. Thresholds developed and evaluated in the same small sample are vulnerable to optimism; balanced case‐control sampling also prevents direct estimation of clinically transportable positive and negative predictive values [30, 32]. The two‐cohort machine‐learning study is a useful incremental‐value test: the worst coronary Z score was clinically meaningful for coronary care but did not materially improve prediction of IVIG resistance [6].

Future studies should report a prespecified outcome, time origin, prediction horizon, complete model coefficients, handling of missing data, internal validation and optimism correction, calibration plots and calibration slope/intercept, overall error measures, and geographically independent external validation. Incremental evaluation should compare candidate measures with a defined clinical‐plus‐basic‐echo model and include change in discrimination, calibration, likelihood ratios where appropriate, and decision‐curve net benefit. An impact study is then needed to determine whether model‐guided monitoring or treatment changes improve patient‐important outcomes without unacceptable over‐treatment.

6.2. Measurement Standardization and Serial‐Data Design

Measurement reliability is a precondition for transportability. The poor inter‐reader reliability of PVB demonstrates that a plausible sign can fail at the level of reproducible acquisition or interpretation [35, 36]. Pediatric quantification standards, the construction of different coronary reference equations, and direct KD variability studies show that acquisition, body‐size modelling, and formula selection all matter [13, 14, 15, 16, 19]. Coronary Z‐score models can systematically reclassify the same measurement [17, 18]. Strain and myocardial‐work values are influenced by vendor, software, frame rate, age, blood pressure, and loading conditions [20, 38, 39, 40]. Future multicenter protocols should define acquisition views, coronary landmarks, cardiac‐cycle timing, body‐surface‐area equation, Z‐score formula, image‐quality criteria, core‐laboratory procedures, and intra‐ and inter‐reader reproducibility.

Serial data require an explicitly longitudinal design. Repeated examinations from one child cannot be treated as independent observations. Studies should prespecify landmarks such as diagnosis, immediately before treatment, early post‐treatment, subacute follow‐up, and convalescence; model both the current value and trajectory; and address informative testing, because sicker children undergo more frequent imaging. Dynamic prediction or joint longitudinal‐outcome models may be more appropriate than a static baseline score.

6.3. Artificial Intelligence, Radiomics, and Three‐Dimensional Imaging

Artificial intelligence could improve coronary border detection, Z‐score calculation, view quality control, and automated strain analysis, but the current KD studies are proofs of concept. A deep‐learning classifier trained on coronary images from 138 children with KD and 65 with pneumonia achieved 84.1% sensitivity but only 58.5% specificity and did not undergo multicenter external validation [58]. A 2025 study quantified coronary brightness in 89 acute KD examinations and used unsupervised clustering to identify a subgroup with larger coronary dimensions and higher bilirubin and pentraxin‐3 concentrations [59]. The latter converts a subjective sign into a reproducible candidate feature, but it neither directly predicts IVIG non‐response nor demonstrates improved outcomes. A targeted PubMed radiomics search through the review cutoff identified no KD‐specific original clinical radiomics study. Representative pediatric datasets, equipment‐domain harmonization, transparent reference standards, calibration, external validation, uncertainty reporting, and clinical‐utility testing remain prerequisites.

Three‐dimensional coronary imaging is similarly attractive but premature. Future studies should compare three‐dimensional aneurysm volume and shape with standardized two‐dimensional dimensions and cross‐sectional angiography, report feasibility and reproducibility, and test whether 3D information changes thrombosis or ischemia risk classification. Until then, the direct KD 3D evidence should not be portrayed as validation of coronary aneurysm morphology analysis [37].

6.4. Limitations of this Review

This is a transparent narrative review rather than a systematic review. Theme‐based searches were designed to support a clinically coherent argument and may have missed eligible reports. Embase, Web of Science/Scopus, CNKI, and Wanfang were not included in the present search, and the synthesis was not based on duplicate independent screening. Publication bias and selective reporting of statistically favorable imaging thresholds are plausible. Several influential studies were small, retrospective, or based on indirect outcomes. The proposed framework is author derived and has not undergone calibration, external validation, impact analysis, or cost‐effectiveness assessment. These limitations are reasons to calibrate the strength of recommendations, not to dismiss the established value of standardized conventional echocardiography.

7. Conclusion

Echocardiography in IVIG‐resistant or high‐risk KD is best understood as a lifecycle decision‐support platform. In the acute phase it enriches, but does not settle, prediction of treatment resistance; during and after treatment it distinguishes clinical defervescence from coronary and myocardial trajectory; in long‐term care it anchors surveillance to the maximum historical coronary burden. Standardized coronary measurement, one consistently applied Z‐score system, conventional ventricular evaluation, and serial interpretation remain the mature core. Strain, myocardial work, RV–PA coupling, CFR, stress imaging, CMR, CTA, and IVUS refine selected phenotypes or resolve specific questions, but none of the candidate advanced indices currently has sufficient validation to direct therapy in isolation. The next advance will depend less on adding another index than on validating a standardized longitudinal dataset against explicitly separated outcomes and demonstrating that imaging‐guided decisions improve patient‐important results.

Funding

This research was supported by the Lanzhou Municipal Science and technology Program (2023‐4‐25).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting information: echo70589‐sup‐0001‐SuppMat.docx

Supporting information: echo70589‐sup‐0002‐TableS1.docx

Data Availability Statement

No new datasets were generated. Search strings and the evidence map are provided in the supplementary materials.