Fig. 2: Quantitative and qualitative segmentation examples on BCSS and PanNuke segmentation datasets. | npj Digital Medicine

Fig. 2: Quantitative and qualitative segmentation examples on BCSS and PanNuke segmentation datasets.

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

Fig. 2

a Comparison of segmentation results with U-Net and FPN on BCSS. b Comparison of segmentation results with U-Net and FPN on PanNuke. Our Pathology-NAS outperforms other two representative methods on both datasets. c Comparison of visualized segmentation examples on BCSS, including lung and breast tissues. d Comparison of visualized segmentation examples on PanNuke, including skin and bile duct tissues. It can be observed from segmentation examples that Pathology-NAS most closely resemble the ground truth, accurately delineating the distribution of different anatomical structures. Our method not only fully covers the target regions that need to be labeled but also meticulously distinguishes the boundaries between the target regions and the background regions.

Back to article page