Fig. 4: The molecular classifier based on machine learning.

A multistep molecular discrimination of 197 SBCLN cases in the validation cohort is shown in a heatmap based on digital expression data. Each column represents an SBCLN case, and each row represents a variable gene in the heatmap (red: high expression, green: low expression; scaled by z statistics). In each row, the gene expression of the remaining cases was scaled by z statistics. All selected genes were included in the refined subset and clustered according to their corresponding entities. Tumor cell content, pathological diagnosis, predictive subgroup by molecular classifier, and integrated diagnosis are indicated above.