Table 2 Comparison of the performance of different classification pipelines as measured by AUC

From: Segmentation-free pretherapeutic assessment of BRAF-status in pediatric low-grade gliomas

 

AUC (95% confidence interval)

Semiautomatic whole tumor volume segmentation

0.756 (0.735, 0.778)

Whole image (Random initialization)

0.747 (0.719, 0.774)

Whole image (Transfer learning)

0.735 (0.704, 0.746)

Whole image (Pretraining)

0.779 (0.759, 0.799)

  1. The best-performing pipeline, in bold, was the whole image model that was pretrained using the semiautomatic whole tumor volume segmentation, radiomic features, and tumor location, before being fine-tuned for pLGG genetic status identification.
Back to article page