摘要
目的构建影像组学模型,提高~(18)F-PSMA-1007 PET/CT对局灶性前列腺癌的鉴别能力。方法回顾性收集2020年7月至2024年4月西安交通大学第一附属医院经前列腺穿刺活检确诊前列腺癌、~(18)F-PSMA-1007 PET/CT显示局灶性核素浓聚的74例患者,中位年龄为71岁,42例患者Gleason评分<8,32例患者Gleason评分≥8,根据检查时间随机选出外部验证集,并将其余患者按照7∶3随机划分为训练集和测试集。在配准图像上半自动勾画病灶感兴趣区(region of interest,ROI)并手动调整,随后将ROI对称性移至对侧非肿瘤组织,通过方差和相关性分析筛选特征,使用Logistic回归构建模型,并与视觉判断比较,通过绘制受试者工作特征(receiver operating characteristic,ROC)曲线比较各个模型的效果。基于Gleason评分、血清tPSA水平和病灶位置划分亚组,获得肿瘤组织鉴别诊断的最佳特征。结果共筛选出8个特征,视觉评价、测试集和外部验证集中曲线下面积(area under the curve,AUC)分别为0.858、0.933和0.891,鉴别前列腺肿瘤的灵敏度分别为0.757、0.800和0.917,特异度分别为0.960、0.800和0.792。亚组分析结果显示,组学特征10percentile和skewness在肿瘤鉴别中价值较高,肿瘤组织10Percentile各组均高于非肿瘤组织(P值分别为0.012、0.002、<0.001、<0.001、<0.001、<0.001);当tPSA≤10 ng/mL且Gleason评分≥8时,肿瘤与非肿瘤组织skewness无统计学差异(P=0.08);tPSA≥20 ng/mL时,非肿瘤组织的skewness稍高于肿瘤组织,但无统计学差异(P值分别为0.285、0.791);肿瘤灶位于前列腺后部(左后部、右后部)时,skewness显著高于非肿瘤组织(P均<0.001)。结论影像组学模型在区分局灶性前列腺癌肿瘤与非肿瘤组织方面灵敏度和准确性优于视觉评价,但后者特异度更高;其中skewness和10percentile对于鉴别诊断有较高价值。
Objective To construct a radiomics model to improve the discriminatory ability of 18 F-PSMA-1007 PET/CT for focal prostate cancer.Methods We retrospectively collected data from 74 patients diagnosed with prostate cancer by biopsy at The First Affiliated Hospital of Xi’an Jiaotong University between July 2020 and April 2024.These patients had focal radionuclide accumulation observed on 18 F-PSMA-1007 PET/CT,with the median age of 71 years.Among them,42 patients had a Gleason score<8 and 32 patients had a Gleason score≥8.An external validation set was randomly selected based on the timing of examination,while the remaining patients were randomly divided into training and test sets at a 7∶3 ratio.Region of interest(ROI)were semi-automatically drawn on registered images,manually adjusted,and symmetrically shifted to contralateral non-tumor tissue.We made variance and correlation analyses to choose features,and built models with Logistic regression and compared the results with those of visual evaluation.Receiver operating characteristics(ROC)curves were drawn to compare model performance,and subgroup analysis was performed to identify optimal features for distinguishing tumor tissue,based on Gleason score,serum total prostate specific antigen(tPSA)levels,and lesion location.Results A total of eight features were selected.The area under the curve(AUC)for visual evaluation,testing set,and external validation set were 0.858,0.933,and 0.891,respectively.The sensitivity was 0.757,0.800 and 0.917;the specificity was 0.960,0.800 and 0.792,respectively.Subgroup analysis showed that the radiomic features 10percentile and skewness had a high value in tumor differentiation.In tumor tissues,the 10percentile values were higher than in non-tumor tissues across all groups(P-values were 0.012,0.002,<0.001,<0.001,<0.001,and<0.001).When tPSA≤10 ng/mL and Gleason score≥8,there was no statistically significant difference in skewness between tumor and non-tumor tissues(P=0.08).When tPSA≥20 ng/mL,the skewness of non-tumor tissue was slightly higher than that of tumor tissue,but the difference was not statistically significant(P-values were 0.285 and 0.791).When the tumor was located in the posterior part of the prostate(left posterior and right posterior),the skewness was significantly higher in tumor tissue than in non-tumor tissue(P-values<0.001 for both).Conclusion The radiomics model had better sensitivity and accuracy than visual evaluation in distinguishing focal prostate cancer tumors from non-tumor tissues,but visual evaluation had higher specificity.Skewness and 10percentile had a high value in differential diagnosis.
作者
常儒玺
罗量
王睿妍
董伟璇
段小艺
CHANG Ruxi;LUO Liang;WANG Ruiyan;DONG Weixuan;DUAN Xiaoyi(Department of ET/CT,The First Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710061,China)
出处
《西安交通大学学报(医学版)》
北大核心
2025年第2期339-344,共6页
Journal of Xi’an Jiaotong University(Medical Sciences)
