Figure 1
From: Imaging-based representation and stratification of intra-tumor heterogeneity via tree-edit distance
Patient representation pipeline: lesions’ radiomic vectors of each patient are dimensionally reduced according to view-aware Principal Component Analysis. [Step 1] Features are grouped according to the six semantic group, or view, they are semantically divided into. As to preserve a balanced importance between views, two principal components are kept from the scores of each PCA, leading to different percentages of explained variability. A total of twelve principal components results from the process, which include six orthogonal pairs of linear combinations of original features. [Step 2] Accordingly, patients are represented as finite sets of \(n_i\) points in \({\mathbb {R}}^{12}\), that is the reduced radiomic space according to view-aware strategy implementation. In the example, \(n_i=7\). [Step 3] Pairwise (Euclidean) distance is computed among patients’ lesions and [Step 4] hierarchical clustering with average linkage is applied to distance matrices, resulting in a dendrogram T representing each patient.
