Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

PAS-stained image (left) and IF image (right). The PAS-stained image shows a clear texture, making it easy to see the boundary of glomeruli. However, in the IF image, the texture of the glomeruli appears very similar to the background, which makes it significantly more challenging to distinguish the boundary.

More »

Fig 1 Expand

Fig 2.

Demonstration of sample diversity. (a-d) Positive lesion glomeruli with various appearances. (e-f) Due to staining contamination or slice location influence, it is difficult to tell if it is a glomeruli. (g-h) Negative glomeruli mixed with background. (i) Background similar to glomeruli.

More »

Fig 2 Expand

Fig 3.

Schematic diagram of the dataset production process. The original WSIs are cropped on magnification 10× and 20× after doctor labeling to obtain the final dataset.

More »

Fig 3 Expand

Table 1.

Distribution of data sets.

More »

Table 1 Expand

Fig 4.

Scale and number of glomeruli in patches from different magnifications.

More »

Fig 4 Expand

Fig 5.

The overall architecture of GlomSAM.

More »

Fig 5 Expand

Fig 6.

Schematic diagram of prompt generator. a: SAM’s ViT Block. b: Our improved ViT Block. c: Details of the Prompt Generator module.

More »

Fig 6 Expand

Fig 7.

Rough mask generator.

More »

Fig 7 Expand

Table 2.

Test results of different models on 10× images (Rounded to two decimal places).

More »

Table 2 Expand

Fig 8.

Distribution of model metrics on 10× images. From the distribution graph, the segmentation effect of each model can be compared intuitively. It can be seen that the segmentation performance of the traditional models is unstable while the SAM series models have excellent stability.

More »

Fig 8 Expand

Table 3.

Test results of different models on 20× images (rounded to two decimal places).

More »

Table 3 Expand

Fig 9.

Distribution of model metrics on 20× image. It shows evident improvement in the performance of all models compared to 10× images, and most of the models have stable performance.

More »

Fig 9 Expand

Fig 10.

Sample diagram of the SAM series models. Here, we mark the raw image with different color areas. The yellow area in the image indicates the ground truth in the first column and the segmentation results of models in other columns. The red box marker indicates that the model cannot segment the accurate boundary. The blue box marker cannot segment the boundary in a detailed and rounded way with a jagged shape.

More »

Fig 10 Expand

Fig 11.

Qualitative comparison results between GlomSAM and other models.

More »

Fig 11 Expand

Table 4.

Ablation study results.

More »

Table 4 Expand