摘要
目的探究能谱CT低剂量图像对影像组学模型鉴别结直肠癌(CRC)分化程度的效能影响。方法常规剂量组(A组)和低剂量组(B组)均纳入行能谱CT全腹增强扫描且经病理证实为结直肠腺癌(CRAC)的患者,并根据病理结果标记为分化良好或分化不良。采用倾向性评分匹配法(PSM)平衡两组基线特征,最终两组各纳入患者111例。A组管电压80/140 kVp瞬切,管电流固定400 mA;B组管电压80/140 kVp瞬切,管电流采用GSI-Assist模式,噪声指数(NI)设为12。A组重组70 keV结合后置40%自适应统计迭代算法(ASIR-V)动、静脉期图像,B组重组70keV结合后置60%ASIR-V的动、静脉期图像,扫描完成后计算有效辐射剂量(ERD)。两组患者分别按照7∶3的比例分为训练集和测试集,并勾画动、静脉期CRC的体积感兴趣区(VOI)。通过Mann-Whitney U检验,斯皮尔曼相关系数(SCC)和LASSO筛选出最有价值的特征集。利用极端随机树(Extra Trees)分类器构建6个预测CRAC分化程度的影像组学模型,并比较模型效能。结果经过倾向性评分匹配(PSM)配对后两组患者一般资料差异无统计学意义(P>0.05)。B组ERD相较于A组降低31.43%(P<0.05)。在测试集中,A组动静脉期联合模型的诊断效能最高,其曲线下面积(AUC)值(0.824)大于A组动脉期模型(0.723)和A组静脉期模型(0.766)。在测试集中,B组动静脉期联合模型的诊断效能最高,其AUC值(0.830)大于B组动脉期模型(0.697)和B组静脉期模型(0.750)。B组静脉期和动静脉期联合模型的AUC与A组相当(P=0.902、0.989)。结论能谱CT低剂量图像不会对影像组学模型在鉴别CRAC分化程度方面产生负面影响,可以在保持诊断效能的同时降低患者的辐射剂量,为CRC患者术前无创鉴别分化程度提供了一种安全和高效的选择。
Objective To explore the effect of spectral CT low dose images on the diagnostic performance of radiomics models in identifying colorectal cancer(CRC)differentiation degrees.Methods The conventional dose group(group A)and the low dose group(group B)both included patients who underwent spectral CT abdominal contrast-enhanced examinations and were confirmed by pathological exams as colorectal adenocarcinoma(CRAC).According to the pathological results,the patients were labeled as well-differentiated and poorly differentiated.Propensity score matching(PSM)was applied to balance the differences in the baseline characteristics among the two groups.Finally,groups A and B respectively included 111 patients.Group A used tube voltage of 80/140 kVp,and fixed tube current(400 mA).Group B used tube voltage of 80/140 kVp,and tube current of GSI-Assist mode,with noise index(NI)of 12.Group A was reconstructed at 70 keV combined with post-set 40%adaptive statistical iterative reconstruction-Veo(ASIR-V)to obtain arterial and venous phase images.Group B was reconstructed at 70 keV combined with post-set 60%ASIR-V to obtain arterial and venous phase images.After scanning,the effective radiation dose(ERD)was calculated.The patients in two groups were divided into a training cohort and a test cohort at a ratio of 7∶3,as well as delineating the volume of interest(VOI)of CRC in the arterial phase and venous phase.The most valuable features were selected through Mann⁃Whitney U test,Spearman correlation coefficient(SCC),and least absolute shrinkage and selection operator(LASSO).The Extra Trees classifier was applied to construct six radiomics models for predicting CRAC differentiation degrees,as well as comparing model performance.Results There was no statistically significant difference in baseline characteristics between the two groups after PSM(P>0.05).The ERD of group B decreased by 31.43%compared with group A(P<0.05).In the test cohort,the combined arterial and venous phases model of group A had the best diagnostic performance,with the area under the curve(AUC)value(0.824)higher than that of the arterial phase model of group A(0.723)and the venous phase model of group A(0.766).In the test cohort,the combined arterial and venous phases model of group B had the best diagnostic performance,with an AUC(0.830)higher than that of the arterial phase model of group B(0.697)and the venous phase model of group B(0.750).The venous phase model of group B and the combined arterial and venous phases model of group B both had comparable AUCs with those of group A(P=0.902,0.989).Conclusion There is no negative impact on discriminating CRAC differentiation degree based on spectral CT low dose images,which can reduce the radiation dose while maintaining the diagnostic performance,serving as a safe and high-efficiency option for CRC patients to realize preoperative non-invasive discrimination of CRC differentiation degree.
作者
张竞颐
王诗耕
魏巍
杜长谕
何健
刘义军
魏强
ZHANG Jingyi;WANG Shigeng;WEI Wei(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian,Liaoning Province 116011,P.R.China)
出处
《临床放射学杂志》
北大核心
2025年第9期1684-1692,共9页
Journal of Clinical Radiology
