Blur Detection

🎭 Created Masks

1. 📍 PIQE Activity Mask (per-pixel)

   Binary mask composed of 16x16 px blocks marking blurred areas. Note that the produced masks are downsampled compared to the original WSI.

   🏷️ Block values semantics

   The computed binary masks are scaled into the \([0, 255]\) range to allow for easier visualization. Therefore, the pixel values have the following semantics:

   • \(0\hphantom{55} \rightarrow\) No Artifact
   • \(255 \rightarrow\) Artifact

   However, due to downsampling of the created masks using local mean, it is safer to assume that any non-zero value indicates an artifact.

   Format of the Mask Name

   Piqe_piqe_median_activity_mask_<SLIDE_NAME>.tiff

2. 🧱 PIQE Median (per-tile)

   Blur score mask, computed as a mean value of corresponding activity_mask blocks. Only foreground pixels are taken into account. Note that the produced masks are downsampled compared to the original WSI. Originally, the masks contain floating point values from the \([0.0, 1.0]\) range. These values are then scaled into the \([0, 255]\) range.

   🏷️ Pixel values semantics

   • \(0\hphantom{55} \rightarrow\) Perfectly focused
   • \(255 \rightarrow\) Completely blurred

   Format of the Mask Name

   Piqe_focus_score_piqe_median_<SLIDE_NAME>.tiff

🏗️ Current Limitations

Blocks with value \(255\)

Activity mask blocks may be \(255\) not only for globally blurred regions, but also for locally defocused or sparse areas (e.g. tissue gaps or empty regions). This means a value of \(255\) does not always imply a complete blurr.