Embedded Files
Protocol for Creating Ground Truth in MicroSAM
Prepare your data:
Collect images for each timepoint (e.g., t0, t48, t95, t143) for each channel (T-cells and organoids separately).
Ensure you have:
-A set of T-cell images for each timepoint.
-A set of organoid images for each timepoint.
Organize your data:
Create separate folders for each timepoint and channel.
Upload images to MicroSAM:
Annotate images:
Go to Plugins > SegmentAnything for Microscopy > Annotator 3D in MicroSAM.
Choose the model to use for segmentation.
Click on “Select Directory” to specify where the embeddings will be saved.
Click on “Compute Embeddings” to begin the process of embedding.
Automatic Segmentation:
After embeddings are computed, click on Automatic Segmentation to generate initial segmentations of the objects.
Manual Correction (if necessary):
After automatic segmentation, manually any errors in the segmentation:
To remove unwanted surfaces, use the syringe tool to click on the background and select the surface to delete.
To create new annotations, use prompts like rectangles or points, then click on the cell of interest to apply the changes.
Apply changes across all Z slices by clicking “Segment All Slices” to ensure consistency across all layers of the image.
Export annotations:
To export the image with labels in image format, follow these steps:
Select File > Save Selected Layers.
Choose the export format:
“napari builtins labels (*.tif *.tiff *.bsdf. *im .*lsm .*npz .*pbm .*pcx .*pgm .*ppm .*stk)”.
Use the annotations for model training:
Once the ground truth annotations are ready and exported, use them to train your model.
These annotations will help the model learn how to segment and track changes over time.
Page updated
Google Sites
Report abuse