摘要
目的:通过提取前列腺不同病理Gleason评分患者的MR ADC图和T2WI-SPAIR图像的影像组学特征,构建影像组学模型,比较其对临床显著性前列腺癌的诊断效能.方法:通过提取天水市第一人民医院 2021 年 1 月—2023 年 10 月收治的 93 例前列腺癌患者(其中csPCa 66例,ciPCa 27例)MR ADC图和T2WI-SPAIR图像中病灶影像纹理等组学特征,进行降维筛选,分别筛选出7个和9个最佳非零系数特征,运用机器学习支持向量机(SVM)分别构建ADC图和T2WI-SPAIR诊断临床显著性前列腺癌的影像组学模型,并对其诊断效能进行比较.结果:两种模型训练组和测试组的AUC值分别为 0.882、0.888 和 0.881、0.694,特异度分别为 80.4%、84.8%和85.0%、65.0%;灵敏度分别为89.5%、89.5%、87.5%、75.0%.结论:基于ADC图和T2WI-SPAIR影像组学模型,对临床显著性前列腺癌诊断效能而言,ADC图较T2WI-SPAIR更趋于稳定,但两者诊断临床显著性前列腺癌的精确度差别小.
Abstract
Objective The imaging omics model was constructed by extracting the imaging omics features of ADC and T2WI-SPAIR images of MRI for patients with different Gleason scores in prostate cancer(PCa)pathology,and to compare its diagnostic efficacy for clinically significant prostate cancer.Methods By extracting omics features of lesion texture from ADC map and T2WI-SPAIR images of 93 prostate cancer patients admitted to the Tianshui First People's Hospital from January 2021 to October 2023(including clinically significant PCa 66 cases and insignificant PCa 27 cases),dimensionality reduction screening was performed to select 7 and 9 optimal non-zero coefficient features,respectively.Machine learning support vector machine(SVM)was used to construct imaging omics models for diagnostic clinically significant prostate cancer,and appraised their efficiency.Results The AUC values of the two model training and testing cohort were 0.882,0.888,and 0.881,0.694,respectively.The specificity of training and testing cohort were 80.4%,84.8%,and 85.0%,65.0%,respectively.The sensitivity of training and testing cohort were 89.5%,89.5%,and 87.5%,75.0%,respectively.Conclusion Based on the ADC map and T2WI-SPAIR imaging omics model,the diagnostic efficacy of ADC map for clinically significant prostate cancer tends to be more stable than T2WI-SPAIR,but the accuracy difference in diagnosing clinically significant prostate cancer between the two is small.
万方数据