Decision Limits for Fetal Speckle-Tracking Outputs: Minimal Detectable Change and Threshold Instability in FetalHQ

Background: Fetal Heart Quantification (FetalHQ) enables detailed assessment of fetal cardiac morphology and function using two-dimensional speckle-tracking echocardiography. However, previous studies have focused primarily on conventional reproducibility metrics such as the intraclass correlation coefficient (ICC). From a clinical perspective, it is more relevant to determine what magnitude of change exceeds measurement noise and whether a single result may lead to unstable classification relative to normative thresholds. The aim of this study was to evaluate the decision limits of selected FetalHQ parameters, including the minimal detectable change (MDC95), threshold instability, and the theoretical effect of averaging two independent analyses. Materials and Methods: This methodological study included 93 paired analyses (_1 and _2) obtained from the same recorded four-chamber cine loop under the assumption of zero biological variability. The intra-operator subset (n = 34) and inter-operator subset (n = 59) were analyzed separately. For key parameters, we determined the proportion of usable pairs after quality filtering, bias, Bland-Altman 95% limits of agreement, MDC95, classification flip rates for normalized outputs, and the width of the uncertainty zone (grey zone) around diagnostic thresholds. In addition, we estimated the effect of a mean-of-two workflow on measurement error reduction. Results: The best analytical performance was observed for left ventricular volumetric parameters, particularly LV EDV and LV ESV, which showed the highest proportion of usable pairs (88.2-94.9%) and relatively low MDC95 values. Functional and deformation metrics performed substantially worse, including LV EF, LV FRAC, LV GLS, and RV TAPSE, all of which were characterized by lower usability and wider limits of agreement, especially in the inter-operator analysis. Classification instability was greatest for normalized parameters. In the inter-operator analysis, classification flips occurred in 33.3% of cases for LV EF centile <5, 31.4% for RV TAPSE lateral centile <5, 35.7% for LV GLS centile >95, and 39.2% for LV GLS Z-score >2. Under the theoretical assumption of independent random measurement error, the estimated MDC95 for the mean of two analyses was approximately 29% lower than that for a single analysis; this reduction was mathematically derived rather than prospectively observed. Conclusions: FetalHQ showed the highest analytical stability for simpler LV volumetric parameters, whereas functional and deformation outputs remained substantially more susceptible to measurement noise and classification instability, particularly between operators. In practice, this means that a single measurement obtained near normative thresholds may be insufficient for reliable interpretation. A theoretical mean-of-two approach may reduce random measurement uncertainty, but this strategy was not prospectively validated and does not overcome the limitations of the least analytically stable parameters.