Figure 1.

Workflow of quantitative DCE-MRI analysis. (A) The input consists of DCE-MRI and contours of the prostate, peripheral zone (PZ), and a sample of gluteus maximum (GM); (B) Motion correction of the prostate; (C) Non negative matrix factorization (NMF). The data are presented as a product of three temporal patterns (S₁, S₂, S₃) and their magnitudes (W₁, W₂, W₃). The well perfused pattern S_wp is identified between the three patterns as the pattern with the largest AUC for the first 90 s; its corresponding weights W_wp represent an intensity map of the distribution of the well-perfused pixels in the data; (D) W_wp is segmented to identify the suspected for tumor region of interest ROI_wp; (E) ROI_wp is assigned to PZ if >10% of ROI_wp is within the PZ contour and vice versa; (F,G) Series of quantitative features (DCE-score) are computed using the signal-vs-time S_ROI from ROI_wp and S_GM of GM; (H) A spatial map of tumor aggressiveness is computed using DCE-score and W_wp.