create_heatmaps.py [-h] -m MODEL_PATH -o OUTPUT_PATH -t TRUE_CLASS
[--no-pool]
[--mask-threshold THRESH]
[--att-upper-threshold THRESH]
[--att-lower-threshold THRESH]
[--score-threshold THRESH]
[--att-cmap CMAP]
[--score-cmap CMAP]
SLIDE [SLIDE ...]Create heatmaps for MIL models.
| Positional Arguments | Description |
|---|---|
SLIDE |
Slides to create heatmaps for. If multiple slides are given, the normalization of the attention / score maps' intensities will be performed across all slides. |
| Options | Description |
|---|---|
-m MODEL_PATH, --model-path MODEL_PATH |
MIL model used to generate attention / score maps. |
-o OUTPUT_PATH, --output-path OUTPUT_PATH |
Path to save results to. |
-t TRUE_CLASS, --true-class TRUE_CLASS |
Class to be rendered as "hot" in the heatmap. |
--no-pool |
Do not average pool features after feature extraction phase. |
--cache-dir CACHE_DIR |
Directory to cache extracted features etc. in. |
| Thresholds | Description |
|---|---|
--mask-threshold THRESH |
Brightness threshold for background removal. |
--att-upper-threshold THRESH |
Quantile to squash attention from during attention scaling (e.g. 0.99 will lead to the top 1% of attention scores to become 1) |
--att-lower-threshold THRESH |
Quantile to squash attention to during attention scaling (e.g. 0.01 will lead to the bottom 1% of attention scores to become 0) |
--score-threshold THRESH |
Quantile to consider in score scaling (e.g. 0.95 will discard the top / bottom 5% of score values as outliers) |
| Colors | Description |
|---|---|
--att-cmap CMAP |
Color map to use for the attention heatmap. |
--score-cmap CMAP |
Color map to use for the score heatmap. |
The heatmap script can be conveniently run in a podman container. To do so, use
the heatmaps-container.sh convenience script.
./heatmaps-container.sh \
-t TRUE_CLASS \
/wsis/slide1.svs \
/wsis/slide2.svsIn order to use GPU acceleration, the nvidia-container-toolkit has to be
installed beforehand.