Merge pull request #127 from mducducd/fix/supertile-size

Fix/ compute supertile size
The supertile size formula (micron) derives the size from slide pixels * MPP. It can be affected by rounding (np.ceil) and slight metadata inaccuracies in slide_mpp. In some cases, this leads to off-by-one errors or out-of-bounds issues when reading regions. It can be fixed by:

- If the slide’s MPP metadata is available, use the pixel-based formula.
- If mpp metadata is missing and an approximated mpp is used instead (may cause inaccurate or inconsistent), use the tile-based formula (tile micron * num_tiles in supertile) to avoid rounding errors and out-of-bounds issues.

Author: mducducd

src/stamp/preprocessing/tiling.py

@@ -314,6 +314,11 @@ def _supertiles(
     )
     supertile_size_tile_px = TilePixels(tile_size_px * len_of_supertile_in_tiles)
 
+    if default_slide_mpp is not None:
+        supertile_size_um = Microns(tile_size_um * len_of_supertile_in_tiles)
+    else:
+        supertile_size_um = Microns(supertile_size_slide_px * slide_mpp)
+
     with futures.ThreadPoolExecutor(max_workers) as executor:
         futs = []
         for coords_slide_px in _foreground_coords(
@@ -334,7 +339,7 @@ def _supertiles(
                         x=Microns(x_slide_px * slide_mpp),
                         y=Microns(y_slide_px * slide_mpp),
                     ),
-                    size=Microns(supertile_size_slide_px * slide_mpp),
+                    size=supertile_size_um,
                 ),
                 x_slide_px=coords_slide_px.x,
                 y_slide_px=coords_slide_px.y,