基于深度图像先验网络的^(18)F-FDG PET短时间Patlak参数成像

^(18)F-FDG PET Patlak parametric imaging using deep image prior with a reduced scan time

目的:提出基于深度图像先验网络的方法应用于PET短时间动态扫描Patlak参数成像。方法:利用仿真数据和动态采集60 min数据,分别用20~60 min、30~60 min、40~60 min、44~60 min时间段数据进行Patlak参数成像,使用深度图像先验网络方法对不同时间段数据生成的参数图像进行去噪,根据去噪图像分析短时间动态扫描协议的定量变化。结果:本文提出的方法在仿真实验中既有明显的去噪效果又能够保持图像细节,针对临床数据44~60 min时间段生成的Patlak参数图像与20~60 min结果的偏差(Bias)相差值在去噪前后从15.54%减少到6.3%,变异系数(COV)在44~60 min时去噪前后的值在两个背景区域下降率为162.96%和223.08%。同时短时间动态扫描协议下去噪图像定量偏差小于4%。结论:本文提出的方法可以提升Patlak参数图像质量,且可实现短时间动态扫描参数成像。

Objective:Proposing deep image prior(DIP)denoising method for Patlak parametric imaging with reduced scan time.Methods:Using simulated data and 60min clinical dynamic scanning tumor data,the parametric image was generated by the Patlak method under different duration times(20~60min,30~60min,40~60min,44~60min).Then we optimized parametric image quality using deep image prior denoising method and analyzed the quantitative results of different short-duration scan protocols.Results:The simulated experiment results showed the proposed method provided a superior qualitative and quantitative appearance compared with the Gaussian filter and non-local mean filter method.Mean while,the difference of bias between 44~60min and 20~60min duration in clinical data was reduced from 15.54%to 6.3%after image denoising.And the same time,the values of coefficient of variation in 44~60min duration were reduced by 162.96%and 223.08%in two background regions.Simultaneously,the quantitative deviation of denoised images is less than 4%using the short duration dynamic scanning protocol.Conclusion:The DIP method can improve the quality of Patlak parametric images and achieve short-duration dynamic scanning parametric imaging.