Fig. 2

Data flow pipeline for our study. Here, we take a DSA sequence with five frames as an example to illustrate the data processing scheme. For each sequence, we removed every other frame (but always keeping the 1-st frame). The remaining adjacent (e.g., 3rd and 5th) frames are input into a model to predict the missing intermediate frame (i.e., 4th). We use the objective metrics of PSNR, SSIM, and RMSE to quantitatively compare the closeness of the predicted frames to the real (removed) frames. Finally, the remaining (i.e., 1st, 3rd, and 5th) frames and the predicted (i.e., 2nd and 4th) frames are stacked together into a sequence according to priority for the subjective experiments by interventional radiologists