Prospective estimation of water equivalent diameter and SSDE using 3D landmark CT scans

3D Landmark (3DLM) scans have recently been introduced in some new CT scanners for automatic scan planning instead of paired 2D localizers, with potential additional utility.

The purpose of this work was to measure the accuracy and precision of 3DLM scans for determining water equivalent diameter ( $D_w$ ) and size-specific dose estimate (SSDE) in patient CT exams.

A cohort of 223 diagnostic CT series comprised of 82 Head [52 non-contrast (NC); 30 contrast-enhanced (CE)], 66 Chest (36 NC; 30 CE), and 75 Abdomen-pelvis (45 NC; 30 CE) series, acquired on a CT scanner equipped with 3DLM capabilities, were retrospectively collected. All series were acquired for patient standard-of-care, and 3DLM series were generated according to the manufacturer's default protocol. $D_w$ values were measured on the central axial image (along z) of each CT series and 3DLM series, using established methods, and were independently used to derive SSDE. Accuracy and precision of 3DLM series relative to reference NC and CE series were assessed for both $D_w$ and SSDE using Bland-Altman analysis and Wilcoxon signed-rank test across the three body regions. Root mean square error (RMSE) of $D_w$ , computed across series within each protocol, and absolute relative error (ARE) of SSDE, computed per series, were compared against AAPM-established clinical acceptability thresholds (10% per TG-220; 20% per TG-204/293). Proportional bias with patient size ( $D_w$ ) was evaluated using ordinary least squares regression. Per-series differences in $D_w$ and SSDE bias between NC and CE series were assessed within each body region using the Mann-Whitney U test. SSDE determination performance was compared between the new [SSDE(3DLM, $D_w)$ ] and conventional [SSDE(AP/LAT, $D_w$ )] methods using the Wilcoxon signed-rank test to assess differences in bias (accuracy) and Levene's test to assess difference in error variability (precision).

The mean [range] patient age was 45.9 years old [0.1-91.5]. The range of $D_w$ for the reference NC/CE series was 12.6 to 20.9 cm for Head, 9.8 to 35.6 cm for Chest, and 12.7 to 41.6 cm for the Abdomen-pelvis. The $D_w$ RMSE per protocol and SSDE ARE per series were < 3% and < 7.1%, respectively, substantially below AAPM-established clinical acceptability thresholds. A statistically significant bias in $D_w$ and SSDE was identified for the Head and Chest (p < 0.001 for all comparisons) but the median biases were small ( $D_w$ range: -0.6% to -2.1%; SSDE range: (0.4% to 1.9%). Unidirectional $D_w$ underestimation and SSDE overestimation was observed only for the Head protocols. $D_w$ estimation bias showed negligible dependence on patient size (R² < 0.185) across all protocols. $D_w$ and SSDE bias differed significantly between the NC and CE series in the Head (p_Mann-Whitney < 0.001) and Chest (p_Mann-Whitney < 0.05). SSDE(3DLM, $D_w$ ) provided significantly lower median bias compared to SSDE(AP/LAT, $D_w$ ) for the Head and Chest (p < 0.001), as well as lower per-case variability in error across all protocols (p_Levene < 0.001).

3DLM-based measurements of $D_w$ and SSDE achieved clinically robust accuracy and precision relative to the reference standard, with SSDE(3DLM, $D_w$ ) also outperforming the conventional AP/LAT-derived method.