Fig. 4

Testing of the LSTM using denoised dynamic images obtained from the a Noise2Noise network (at position c in Fig. 1), and using b the image derived AIF masked using a vasculature atlas (threshold set at > 0.8) to identify highly probable regions for arterial vasculature. After feeding these as input into the pretrained LSTM kinetic parameters are produced, which are used to reconstruct c dynamic images. Parametric images are further processed using d an unsupervised deep image prior and e denoised version of dynamic images based on having denoised individual parametric images