Fig. 3: Individual quantitative and qualitative segmentation examples on the PanNuke dataset. | npj Digital Medicine

Fig. 3: Individual quantitative and qualitative segmentation examples on the PanNuke dataset.

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

Fig. 3

The figure presents Dice score comparisons (bar charts) and visualized segmentation examples (Original, Ground Truth, Pathology-NAS (Ours), U-Net, FPN) for sixteen distinct tissue types. The masks generated by Pathology-NAS (Ours) generally exhibit a closer alignment with the ground truth annotations, particularly in accurately delineating intricate structures and reducing errors such as over-segmentation when compared to U-Net and FPN. This visual assessment is supported by the quantitative Dice scores, where our method often shows improved or competitive performance across the various tissue types. The red boxes are used to highlight common failure cases in the FPN and U-Net results, such as missed segmentation, incorrect segmentation, and excessive adhesion between different targets, areas where Pathology-NAS (Ours) often demonstrates more robust results.

Back to article page