Fig 1. Spearman rank correlation of radiomics features in the training dataset.

Forty-five features were extracted from tumor volumes in eighty-three patients. Across all tumors, the correlation of each feature with all other features was investigated via Spearman’s rank correlation. The color and size of the circle indicate the degree of correlation. The final radiomics features were decided based on Spearman’s correlation coefficient >0.9 and significant p-value >0.01 after Holm-Bonferroni correction.