Fig. 1: Imaging signatures of IDH-wildtype GBM tumors.

Signatures generated using deep learning (DL; subfigure a; n = 357 independent samples), and conventional machine learning (ML) using only conventional MRI (subfigure b; n = 357 independent samples) as well as using both conventional and advanced MRI (subfigure c; n = 228 independent samples) in mutant (red boxes) and wildtype (blue boxes) tumors. These signatures in Panels (a & c) are shown for co-occurring (left) and exclusive (right) mutations and pathways. The numbers indicated at the top of each pair of mutant-wildtype box-whisker plots indicate the Cohen’s D effect sizes for the signatures generated by DL, ML (conventional MRI), ML (conventional and advanced MRI) models for co-occurring and mutually exclusive tumors. Error bars in the box-and-whisker plots represent the range of the data excluding outliers, with whiskers extending to the smallest and largest values within 1.5 times the interquartile range (IQR) from the lower and upper quartiles. The outliers are represented as individual points outside this range. d A representative example of the degree of expression of imaging signatures, quantitatively measured along respective axes, for mutations in the EGFR (dark blue), PTEN (green), and TP53 (yellow) genes, wildtype tumors (red), and tumors with a combination of EGFR, PTEN, and TP53 mutations (light blue) [iEGFR, iPTEN, and iTP53 = imaging signatures of EGFR, PTEN, and TP53 mutations, respectively]. The colors of the ellipsoids represent the molecularly defined mutation status, and the shape of the ellipsoids reflects the spread of values of respective imaging signatures. For example, molecularly defined EGFR mutant tumors (dark blue) tend to express the iEGFR signature, i.e. they have higher values along the iEGFR axis.