Medical Physics Department, Veneto Institute of Oncology IOV - IRCCS, Padua, Italy.
University of Padua, Padua, Italy.
Eur J Radiol. 2023 Jun;163:110804. doi: 10.1016/j.ejrad.2023.110804. Epub 2023 Mar 29.
To establish size-dependent DRL and to estimate the effectiveness of the size-dependent DRLs over size-independent DRLs for a CT exposure optimization process.
The study included 16,933 adult CT body examinations of the most common CT protocols. Acquisitions were included following an image quality assessment. Patients were grouped into five different classes by means of the water equivalent diameter (D): 21 ≤ D < 25, 25 ≤ D < 29, 29 ≤ D < 33,33 ≤ D < 37 (in cm). CTDI, DLP, DLP and SSDE median values were provided both for the sample as a whole (size-independent approach) and for each D class (size-dependent approach). The performance of the two approaches in classifying sub-optimal examinations was evaluated through the confusion matrix and Matthews Correlation Coefficient (MCC) metric. The 75th percentile of the CTDI distribution was arbitrarily chosen as a threshold level above which the acquisitions are considered sub-optimal.
CTDI, DLP, DLP and SSDE typical values (median values) are statistically different across D groups. The confusion matrix analysis suggests that size-independent DRL could not mark potential suboptimal protocols for small and large patients. The agreement between the size-dependent and size-independent methods is strong only for the most populous classes (MCC > 0.7). For small and large patients size-independent approach fails to identify as sub-optimal around 20 % of the acquisition (MCC≪0.2).
It was proven by means of the confusion matrix and MCC metric that stratifying DRLs by patient size, size-dependent DRL can be a powerful strategy in order to improve the dose optimization process shown that a size-independent DRL fails to identify sub-optimal examinations for small and large patients.
建立基于大小的 DRL，并评估基于大小的 DRL 在 CT 曝光优化过程中的有效性是否优于独立于大小的 DRL。
本研究纳入了最常见 CT 协议的 16933 例成人 CT 体部检查。在进行图像质量评估后，对采集结果进行了收录。通过水当量直径（D）将患者分为五个不同的组别：21≤D<25、25≤D<29、29≤D<33、33≤D<37（cm）。为整个样本（独立于大小的方法）和每个 D 组（基于大小的方法）提供了 CTDI、DLP、DLP 和 SSDE 的中位数值。通过混淆矩阵和马修斯相关系数（MCC）指标评估两种方法在分类次优检查中的表现。任意选择 CTDI 分布的第 75 个百分位数作为次优采集的阈值水平。
CTDI、DLP、DLP 和 SSDE 的典型值（中位数）在 D 组之间存在统计学差异。混淆矩阵分析表明，独立于大小的 DRL 不能标记小患者和大患者的潜在次优方案。仅对于最常见的组别（MCC>0.7），基于大小的方法和独立于大小的方法之间的一致性才较强。对于小患者和大患者，独立于大小的方法无法识别约 20%的次优采集（MCC≪0.2）。
通过混淆矩阵和 MCC 指标证明，根据患者大小分层 DRL 是一种强有力的策略，可以改善剂量优化过程，表明独立于大小的 DRL 无法识别小患者和大患者的次优检查。
