Department of Radiology, The Second Affiliated Hospital of Kunming Medical University, Kunming, China.
Abdom Radiol (NY). 2024 Jan;49(1):182-191. doi: 10.1007/s00261-023-04053-2. Epub 2023 Oct 31.
To investigate different radiomics models based on single phase and the different phase combinations of radiomics features from 3D tri-phasic CT to distinguish RO from chRCC.
A total of 96 patients (30 RO and 66 chRCC) were enrolled in this study. Radiomics features were extracted from unenhanced phase (UP), corticomedullary phase (CMP), and nephrographic phase (NP) CT images. Feature selection was based on the least absolute shrinkage and selection operator regression (LASSO) method. The selected features were used to develop different radiomics models using logistic regression (LR) analysis, including model 1 (UP), model 2(CMP), model 3(NP), model 4(UP+CMP), model 5(UP+NP), model 6(CMP+NP), and model 7(UP+CMP+NP). The radiomics model demonstrating the highest discrimination performance was utilized to construct the combined model (model 8) with clinical factors. A nomogram based on the model 8 was established. To evaluate the diagnostic performance of the different models, the receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used. Delong's test was utilized to assess the statistical significance of the AUC improvement across the models.
Among the seven radiomics models, model 7 exhibited the highest AUC of 0.84 (95% CI 0.69, 0.99), and model 7 demonstrated a significantly superior AUC compared to the other radiomics models (all P < 0.05). The AUC values of radiomics models based on two phases (model4, mode5, mode6) were greater than the models based on single phase (model1, mode2, mode3) (all P < 0.05). Model 3 illustrated the best performance of the three radiomics models based on single phase with an AUC of 0.76 (95% CI 0.57, 099). Model 6 illustrated the best performance of the three radiomics models based on two-phases combination with an AUC of 0.83 (0.66, 0.99). Model 8 achieved an AUC of 0.93 (95% CI 0.83, 1.00) which is higher than those all radiomics models.
Radiomics models based on combination of radiomics features from UP, CMP, and NP can be a useful and promising technique to differentiate RO from chRCC. Moreover, the model combining clinical factors and radiomics features showed better classification performance to distinguish them.
探讨基于单期和不同期相的影像组学特征的不同放射组学模型，以区分 RO 和 chRCC。
本研究共纳入 96 例患者（30 例 RO 和 66 例 chRCC）。从增强前（UP）、皮质期（CMP）和肾实质期（NP）CT 图像中提取影像组学特征。特征选择基于最小绝对收缩和选择算子回归（LASSO）方法。使用逻辑回归（LR）分析，基于所选特征建立不同的放射组学模型，包括模型 1（UP）、模型 2（CMP）、模型 3（NP）、模型 4（UP+CMP）、模型 5（UP+NP）、模型 6（CMP+NP）和模型 7（UP+CMP+NP）。利用临床因素构建基于模型 7 的组合模型（模型 8），并建立基于该模型的列线图。利用受试者工作特征（ROC）曲线和决策曲线分析（DCA）评估不同模型的诊断性能。采用 Delong 检验评估各模型间 AUC 改善的统计学意义。
在七种放射组学模型中，模型 7 的 AUC 最高，为 0.84（95%CI 0.69，0.99），与其他放射组学模型相比，模型 7 的 AUC 显著更高（均 P<0.05）。基于双期相的放射组学模型（模型 4、模型 5、模型 6）的 AUC 值大于基于单期相的模型（模型 1、模型 2、模型 3）（均 P<0.05）。基于单期相的三种放射组学模型中，模型 3 的 AUC 最佳，为 0.76（95%CI 0.57，0.99）。基于双期相组合的三种放射组学模型中，模型 6 的 AUC 最佳，为 0.83（0.66，0.99）。模型 8 的 AUC 为 0.93（95%CI 0.83，1.00），高于所有放射组学模型。
基于 UP、CMP 和 NP 的影像组学特征组合的放射组学模型可作为一种有用且有前途的技术，用于区分 RO 和 chRCC。此外，结合临床因素和影像组学特征的模型在区分它们时表现出更好的分类性能。
