Fig. 5: Downstream tasks evaluation. | Nature Communications

Fig. 5: Downstream tasks evaluation.

From: Contrast-free identification of glioma blood-brain barrier status via generative diffusion AI and non-contrast MRI

Fig. 5: Downstream tasks evaluation.

a Visual examples of glioma segmentation on the MR-2 dataset (n = 94 samples) obtained by different SOTA methods. In the magnified view, green indicates tumor ground truth, red shows segmentation masks on real or synthetic T1Gd images from our pre-trained glioma segmentation model, and blue marks their overlap, with greater overlap signifying better segmentation. b Performance of glioma grading obtained by the existing five synthesis models and our CBSI on samples from the MR-1 dataset with available glioma grade labels (n = 41 samples). Both overall and grade-specific (Grades 2–4) statistical analyses. The radar chart illustrates classification metrics with the four vertices representing AUC, ACC, SEN, and SPE, the center of the circle indicating 40% for the corresponding metric, and the outermost circle representing 100%.

Back to article page