多参数MRI深度学习人工智能分析联合^(68)Ga-PSMA PET对前列腺癌的诊断价值  

Diagnostic value of multiparametric MRI deep learning artificial intelligence analysis combined with ^(68)Ga-PSMA PET for prostate cancer

作　　者：覃春霞[1] 吕玉虎 代志博 魏思齐 阮伟伟 盖永康 兰晓莉[1] Qin Chunxia;Lyu Yuhu;Dai Zhibo;Wei Siqi;Ruan Weiwei;Gai Yongkang;Lan Xiaoli(Department of Nuclear Medicine,Union Hospital,Tongji Medical College,Huazhong University of Science and Technology,Hubei Province Key Laboratory of Molecular Imaging,Wuhan 430022,China)

机构地区：[1]华中科技大学同济医学院附属协和医院核医学科、分子影像湖北省重点实验室,武汉430022

出　　处：《中华核医学与分子影像杂志》2024年第9期516-521,共6页Chinese Journal of Nuclear Medicine and Molecular Imaging

基　　金：湖北省科技创新团队([2022]72)。

摘　　要：目的评估多参数MRI(mpMRI)深度学习人工智能(AI)分析系统联合^(68)Ga-前列腺特异膜抗原(PSMA)PET对前列腺癌的诊断效能。方法回顾性收集2018年5月至2023年10月间因怀疑或确诊前列腺癌在华中科技大学同济医学院附属协和医院行^(68)Ga-PSMA PET/MR检查的103例患者, 年龄45~85岁, 勾画ROI测量病灶/前列腺SUV_(max), 应用前列腺深度学习AI系统对MR图像进行分析, 获得前列腺及原发灶参数。以病理结果为"金标准", 分析T2加权成像(WI)、弥散WI(DWI)、mpMRI、PET SUV_(max)及PET/MR对前列腺癌的诊断效能。结果 103例患者中, 82例(79.61%)为前列腺癌。PET单模态表现出最佳的特异性[100%(21/21)]、阳性预测值[100%(58/58)]和AUC(0.860, 95%CI: 0.777~0.920)。mpMRI AI分析可提供快速的诊断结果, 其联合PET可提高诊断灵敏度和准确性[PET、mpMRI、两者联合的灵敏度分别为:70.73%(58/82)、86.59%(71/82)、92.68%(76/82);准确性分别为76.70%(79/103)、81.55%(84/103)、86.41%(89/103)];在PET阴性的44例患者中, 加入mpMRI AI分析结果后, 30例获得准确诊断。结论 ^(68)Ga-PSMA PET对前列腺癌具有很好的特异性, mpMRI AI分析省时, 两者联合可提高诊断灵敏度和准确性, 为^(68)Ga-PSMA PET/MR图像分析提供有价值的工具。ObjectiveTo evaluate the diagnostic efficacy of the multiparametric MRI(mpMRI)deep learning artificial intelligence(AI)analysis system combined with ^(68)Ga-prostate specific membrane antigen(PSMA)PET for prostate cancer.MethodsData of 103 patients(age:45-85 years)who underwent ^(68)Ga-PSMA PET/MR at Union Hospital,Tongji Medical College,Huazhong University of Science and Technology from May 2018 to October 2023 for suspected or confirmed prostate cancer were retrospectively collected.ROI was delineated to measure SUV max of primary tumor or prostate,and a deep learning AI system was applied to analyze MR images of the prostate.The diagnostic efficacies of T 2 weighted imaging(WI),diffusion WI(DWI),mpMRI,PET SUV max,and PET/MR for prostate cancer were assessed,with using the pathological results as the gold standard.ResultsAmong 103 patients,82 cases(79.61%)were with prostate cancer.PET unimodality demonstrated the best specificity(100%,21/21),positive predictive value(100%,58/58),and AUC(0.860,95%CI:0.777-0.920).The mpMRI AI analysis provided rapid diagnostic results and the sensitivity and accuracy were improved by combining with PET(sensitivities of PET,mpMRI and the combination of the two were 70.73%(58/82),86.59%(71/82),and 92.68%(76/82),respectively;the accuracies were 76.70%(79/103),81.55%(84/103)and 86.41%(89/103),respectively).Among 44 patients with negative PET,30 patients received an accurate diagnosis when the results of mpMRI AI analysis were added.Conclusions ^(68)Ga-PSMA PET demonstrates good specificity for prostate cancer and mpMRI AI analysis is time-saving.The combined application improves the diagnostic sensitivity and accuracy,which provides a valuable tool for ^(68)Ga-PSMA PET/MR image analysis.

关 键 词：前列腺肿瘤 前列腺特异膜抗原 同位素标记 镓放射性同位素 正电子发射断层显像术 多参数磁共振成像 深度学习 

分 类 号：R737.25[医药卫生—肿瘤]