Kokiri: Random Forest-Based Cohort Comparison and Characterization
Eckelt, K.; Adelberger, P.; Bauer, M. J.; Zichner, T.; Streit, M.
Show abstract
AO_SCPLOWBSTRACTC_SCPLOWWe propose an interactive visual analytics approach to characterizing and comparing patient subgroups (i.e., cohorts). Despite having the same disease and similar demographic characteristics, patients respond differently to therapy. One reason for this is the vast number of variables in the genome that influence a patients outcome. Nevertheless, most existing tools do not offer effective means of identifying the attributes that differ most, or look at them in isolation and thus ignore combinatorial effects. To fill this gap, we present Kokiri, a visual analytics approach that aims to separate cohorts based on user-selected data, ranks attributes by their importance in distinguishing between cohorts, and visualizes cohort overlaps and separability. With our approach, users can additionally characterize the homogeneity and outliers of a cohort. To demonstrate the applicability of our approach, we integrated Kokiri into the Coral cohort analysis tool to compare and characterize lung cancer patient cohorts.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.