摘要
目的:提出一种在镓68(^(68)Ga)前列腺特异性膜抗原(68Ga-PSMA-11)正电子发射断层扫描与磁共振(PET/MR)图像上进行前列腺定位与前列腺癌病灶分割的级联网络,以解决前列腺癌病灶分割中的难题。方法:收集医院收治的125例前列腺疾病患者的PET/MR图像数据,按照8∶2的比例分配训练集和测试集,采用五折交叉验证,分别训练基础网络、MR网络和级联网络3个网络。基础网络和级联网络的二级网络是学习T2加权成像(T_(2)WI)、弥散加权成像(DWI)、表观弥散系数(ADC)和PET的4种模态的语义分割(U-Net)网络;MR网络是仅学习MR的3种模态U-Net网络;级联网络的一级网络Faster-RCNN定位前列腺和精囊腺计算Dice系数值进行模型评估。结果:级联网络的Dice系数为0.76,基础网络和MR网络的Dice系数分别为0.73和0.62,结合分割结果视觉图,级联网络对前列腺癌病灶的检测更加准确。结论:级联网络相对于单个U-Net网络,可以更准确分割^(68)Ga-PSMA-11 PET/MR图像上前列腺癌病灶,为临床对前列腺癌的定位诊断提供便利。
Objective:To propose a cascade network that can locate prostate and segment the lesion of prostate cancer on the images of positron emission tomography and magnetic resonance(PET/MR)of gallium 68 prostate specific membrane antigen(^(68)Ga-PSMA-11)so as to solve the problem of segmenting lesion of prostate cancer.Methods:The data of PET/MR images of 125 patients with positive and negative prostate cancer who admitted to hospital were collected,and they were divided into training set and test set according to 8:2.The 50%cross validation was adopted to respectively train basic network,MR network and cascaded network.The second grade network of basic network and cascaded network was a semantic segmentation(U-Net)network which learned from four kinds of modes included T2 weighted imaging(T_(2)WI),diffusion weighted imaging(DWI),apparent diffusion coefficient(ADC)and positron emission tomography(PET).The MR network was a U-Net that learned only from MR three modes.The first-grade network Faster-RCNN of cascade network calculated the Dice coefficient to assess model after located the prostate and seminal vesicle glands.Results:The Dice coefficients of cascade network,basic network and MR network were respectively 0.76,0.73 and 0.62.Combined with the visual image of the results of segmentation,the result of cascade network detection was more accurately for the lesions of prostate cancer.Conclusion:Compared with single U-Net network,the cascade network can more accurately segment the lesions of prostate cancer in ^(68)Ga-PSMA-11 PET/MR images,which can provide convenience for the clinical localization and diagnosis of prostate cancer.
作者
洪文威
徐磊
李如帅
孟庆乐
杨瑞
毛舟
蒋红兵
HONG Wen-wei;XU Lei;LI Ru-shuai(Department of Nuclear Medicine,Nanjing First Hospital,Nanjing Medical University,Nanjing 210006,China;不详)
出处
《中国医学装备》
2023年第6期18-22,共5页
China Medical Equipment
基金
南京市医学科技发展课题(QRX11033)“南京市卫生青年人才培养工程”
2022年度南京市卫生科技发展专项资金(ZKX22036)“基于HYPER Iterative重建技术对68Ga-DOTANOC PET/CT图像质量研究”
2020年度南京市卫生科技发展专项资金(YKK20104)“^(177)Lu-DOTATOC靶向治疗神经内分泌肿瘤的内照射吸收剂量研究”。
