Figure 5
Physionet datascape. (a) Graph of the datascape on the clinical data measured at the 48th hour. The graph is projected in two dimensions thanks to the Fruchterman–Reingold layout algorithm. The red nodes represent deceased patients, whereas the blue ones are those in remission. (b,c) On the left, visualization of the trajectories of resp. a deceased patient and a patient in remission within the datascape over time. The color of the trajectory evolves from purple (1st hour of hospitalization) to yellow (48th hour of hospitalization). The blue lines represent the geodesic between the 1st and the 48th hour of hospitalization—the corresponding evolution of the ratio of deceased patients in the neighborhood of the studied patients pictured on the right panel. (d) Benchmark on 4-fold model training, comparing the distribution of the AUC of the different models benchmarked. The predictive model based on the datascape is trained with several values for the parameter k.
