Pathology outputs

Pathology analysis returns a compact result summary plus downloadable artifacts, and exactly what you get depends on the workflow you ran. Read the summary on the card first — tile count, tissue coverage, or feature dimension — then download the archive, mask, or feature file for downstream work. See inputs for the parameters that shape these results.

Results by workflow

Tile extraction

• num_tiles — how many tiles survived the tissue and blur filters.
• slide_dimensions — the slide's pixel dimensions.
• Artifact: tiles.zip — an archive of the retained tiles in your chosen output_format.

Each retained tile records its position (x, y), tissue_percentage, and blur_score internally, so the archive contains only tiles that passed both quality filters.

Tissue detection

• tissue_percentage — the fraction of the slide covered by tissue, as a percentage.
• num_regions — how many distinct tissue regions were found.
• regions — bounding boxes (x, y, width, height, area) for each region above the artifact-size threshold.
• image_dimensions — the analyzed image dimensions.
• Artifact: tissue_mask.png — the binary tissue mask (when save_masks is on).

Feature extraction

• num_tiles — tiles from which features were extracted.
• feature_dim — the feature vector length (2048 for ResNet-50).
• mean_features — the mean feature vector across tiles.
• Artifact: features.npy — the per-tile feature matrix (when save_features is on).

Common fields

Every completed run also reports processing_time_seconds and uploads a job manifest artifact that records the full run context.

What the card shows

Results render as an intelligence card (a chart-type card) on the canvas. The preview surfaces the headline metrics the analysis produced, which may include:

• Cells — cell count, when a workflow reports one (primary metric).
• Tissue — the tissue-type classification label.
• Annotations — the number of detected regions.
• Confidence — a prediction-confidence score, color-coded (green above 90%).

Expand the card to see the full summary and to reach the downloadable artifacts.

Using the results

• Tiles feed directly into feature extraction, or into an external classifier.
• Tissue masks and coverage quantify how much usable tissue a slide carries before you invest compute in it.
• Feature matrices support clustering, similarity, and classification across tiles or slides.

Read next