Fig. 5: Significance testing of Top-1 accuracy (Prec@1(%)) for different NAS methods on the BreakHis, Diabetic, BCSS and PanNuKe datasets, utilizing ShuffleNet, ViT and U-Net backbones. | npj Digital Medicine

Fig. 5: Significance testing of Top-1 accuracy (Prec@1(%)) for different NAS methods on the BreakHis, Diabetic, BCSS and PanNuKe datasets, utilizing ShuffleNet, ViT and U-Net backbones.

From: Large language models driven neural architecture search for universal and lightweight disease diagnosis on histopathology slide images

Fig. 5

The x-axis displays the compared search methods, including random (i.e., Random Search), Cream, AF (i.e., AutoFormer), and “ours” (i.e., Pathology-NAS), grouped by dataset and backbone. The y-axis represents the Prec@1(%) performance. Asterisks indicate statistical significance levels assessed by independent two-sample Welch’s t tests: * denotes p < 0.05 (significant), ** denotes p < 0.01 (highly significant), and *** denotes p < 0.001 (very highly significant). Prec@1: Top-1 accuracy. Prec@5: Top-5 accuracy.

Back to article page