diff --git a/rationai/mlkit/data/datasets/slides_tiles_loader.py b/rationai/mlkit/data/datasets/slides_tiles_loader.py
index 7359965..d487ebe 100644
--- a/rationai/mlkit/data/datasets/slides_tiles_loader.py
+++ b/rationai/mlkit/data/datasets/slides_tiles_loader.py
@@ -76,9 +76,9 @@ def _build_tile_index(tiles: HFDataset) -> dict[str | bytes, pa.ListScalar]:
         if len(tiles) == 0:
             return {}
 
-        # 1. Grab the column directly from the underlying PyArrow Table
-        slide_ids = tiles.data.column("slide_id")
-        num_rows = len(slide_ids)
+        # 1. Read slide_id as Arrow data, respecting any prior .filter() / .select().
+        slide_ids = tiles.with_format("arrow")["slide_id"]
+        num_rows = len(tiles)
 
         # 2. Handle the "Large" type conversion
         current_type = slide_ids.type
diff --git a/uv.lock b/uv.lock
index 6c5d097..cb69bbf 100644
--- a/uv.lock
+++ b/uv.lock
@@ -2287,7 +2287,7 @@ dependencies = [
 
 [[package]]
 name = "rationai-mlkit"
-version = "0.4.0"
+version = "0.4.1"
 source = { virtual = "." }
 dependencies = [
     { name = "datasets" },