From 5ee5e038ee05f76ee81f797944d2e49366735274 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 06:23:19 +0000
Subject: [PATCH 1/8] feat: no dropout

---
 configs/model/pgp_full_model.yaml      | 1 +
 configs/model/pgp_model.yaml           | 1 +
 configs/model/virchow2_full_model.yaml | 1 +
 configs/model/virchow2_model.yaml      | 1 +
 4 files changed, 4 insertions(+)

diff --git a/configs/model/pgp_full_model.yaml b/configs/model/pgp_full_model.yaml
index ca8f9673..8db2cef2 100644
--- a/configs/model/pgp_full_model.yaml
+++ b/configs/model/pgp_full_model.yaml
@@ -5,6 +5,7 @@ defaults:
 
 decode_head:
   in_features: 1536
+  dropout: 0.0
 
 _target_: ml.foundation_model.FoundationCarcinomaModel
 lr: ???
diff --git a/configs/model/pgp_model.yaml b/configs/model/pgp_model.yaml
index a6305ae2..66303519 100644
--- a/configs/model/pgp_model.yaml
+++ b/configs/model/pgp_model.yaml
@@ -2,5 +2,6 @@ _target_: ml.embedding_model.EmbeddingCarcinomaModel
 decode_head:
   _target_: ml.modeling.decode_head.BinaryEmbeddingClassifier
   in_features: 1536
+  dropout: 0.0
 lr: ???
 tl_threshold: ???
diff --git a/configs/model/virchow2_full_model.yaml b/configs/model/virchow2_full_model.yaml
index 95a8f91c..3558896c 100644
--- a/configs/model/virchow2_full_model.yaml
+++ b/configs/model/virchow2_full_model.yaml
@@ -5,6 +5,7 @@ defaults:
 
 decode_head:
   in_features: 2560
+  dropout: 0.0
 
 _target_: ml.foundation_model.FoundationCarcinomaModel
 lr: ???
diff --git a/configs/model/virchow2_model.yaml b/configs/model/virchow2_model.yaml
index 643836d1..ec52592c 100644
--- a/configs/model/virchow2_model.yaml
+++ b/configs/model/virchow2_model.yaml
@@ -2,5 +2,6 @@ _target_: ml.embedding_model.EmbeddingCarcinomaModel
 decode_head:
   _target_: ml.modeling.decode_head.BinaryEmbeddingClassifier
   in_features: 2560
+  dropout: 0.0
 lr: ???
 tl_threshold: ???

From 1fede291b3fb209ac7918adfa86dc65e96a07115 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 08:14:22 +0000
Subject: [PATCH 2/8] feat: initial exploration

---
 configs/data/breast/base.yaml              |  6 +++
 configs/data/breast/mmci/b20_24_base.yaml  | 26 ++++++++++++
 configs/data/breast/mmci/b20_24_train.yaml |  7 ++++
 configs/data/breast/mmci/b20_24_val.yaml   |  7 ++++
 configs/data/breast/mmci/base.yaml         | 10 +++++
 configs/data/breast/mmci/stats.yaml        |  2 +
 configs/exploration/breast/b20_24.yaml     | 11 +++++
 configs/exploration/breast/base.yaml       |  6 +++
 exploration/b20_24.py                      | 49 ++++++++++++++++++++++
 9 files changed, 124 insertions(+)
 create mode 100644 configs/data/breast/base.yaml
 create mode 100644 configs/data/breast/mmci/b20_24_base.yaml
 create mode 100644 configs/data/breast/mmci/b20_24_train.yaml
 create mode 100644 configs/data/breast/mmci/b20_24_val.yaml
 create mode 100644 configs/data/breast/mmci/base.yaml
 create mode 100644 configs/data/breast/mmci/stats.yaml
 create mode 100644 configs/exploration/breast/b20_24.yaml
 create mode 100644 configs/exploration/breast/base.yaml
 create mode 100644 exploration/b20_24.py

diff --git a/configs/data/breast/base.yaml b/configs/data/breast/base.yaml
new file mode 100644
index 00000000..8921c69f
--- /dev/null
+++ b/configs/data/breast/base.yaml
@@ -0,0 +1,6 @@
+# @package _global_
+
+carcinoma_artifacts: /mnt/projects/breast_cancer/bc
+
+metadata:
+  experiment_name: Breast Cancer
diff --git a/configs/data/breast/mmci/b20_24_base.yaml b/configs/data/breast/mmci/b20_24_base.yaml
new file mode 100644
index 00000000..641bfb37
--- /dev/null
+++ b/configs/data/breast/mmci/b20_24_base.yaml
@@ -0,0 +1,26 @@
+defaults:
+  - base
+  - _self_
+
+
+qc_masks_uri: "mlflow-artifacts:/61/1fa3adfff18b433782163a29bc9d5c81/artifacts/qc_masks"
+tissue_masks_uri: "mlflow-artifacts:/61/b80d4357ba3b44999896ceda971a1884/artifacts/tissue_masks"
+epithelium_masks_uri: "mlflow-artifacts:/61/0af479b624824d9bae0bc830a5b99f9a/artifacts/epithelium_masks"
+
+
+# were used for filtering
+thresholds:
+  tissue_roi_t: 0.01
+
+  # during the training, epithelium is used as a weak carcinoma annotation, the threshold is risen so that wrong epithelium around corners is removed
+  train_pos_tissue_roi_t: 0.5
+  residual_t: 0.25
+  blur_t: 0.4
+  folding_t: 0.25
+
+  epithelium_roi_t: 0.2
+
+metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_metadata.csv"
+
+
+data_name: "MMCI B20-24"
diff --git a/configs/data/breast/mmci/b20_24_train.yaml b/configs/data/breast/mmci/b20_24_train.yaml
new file mode 100644
index 00000000..81301cd0
--- /dev/null
+++ b/configs/data/breast/mmci/b20_24_train.yaml
@@ -0,0 +1,7 @@
+defaults:
+  - b20_24_base
+  - _self_
+
+metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_train_metadata.csv"
+
+data_name: "MMCI B20-24 Train"
diff --git a/configs/data/breast/mmci/b20_24_val.yaml b/configs/data/breast/mmci/b20_24_val.yaml
new file mode 100644
index 00000000..758cb7c2
--- /dev/null
+++ b/configs/data/breast/mmci/b20_24_val.yaml
@@ -0,0 +1,7 @@
+defaults:
+  - b20_24_base
+  - _self_
+
+metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_val_metadata.csv"
+
+data_name: "MMCI B20-24 Val"
diff --git a/configs/data/breast/mmci/base.yaml b/configs/data/breast/mmci/base.yaml
new file mode 100644
index 00000000..f638ba8f
--- /dev/null
+++ b/configs/data/breast/mmci/base.yaml
@@ -0,0 +1,10 @@
+defaults:
+  - stats
+  - /data/breast/base@_global_
+  - _self_
+
+# TBD -> select a reference slide from train set
+stains:
+  hematoxylin: [0.479, 0.823, 0.305]
+  eosin: [0.156, 0.949, 0.273]
+  residual: [-0.189, -0.242, 0.952]
diff --git a/configs/data/breast/mmci/stats.yaml b/configs/data/breast/mmci/stats.yaml
new file mode 100644
index 00000000..48f1eaaa
--- /dev/null
+++ b/configs/data/breast/mmci/stats.yaml
@@ -0,0 +1,2 @@
+mean: [221.9752, 188.3984, 222.95]
+std: [41.3043, 56.8831, 29.308]
diff --git a/configs/exploration/breast/b20_24.yaml b/configs/exploration/breast/b20_24.yaml
new file mode 100644
index 00000000..9bcc2d07
--- /dev/null
+++ b/configs/exploration/breast/b20_24.yaml
@@ -0,0 +1,11 @@
+# @package _global_
+
+defaults:
+  - /exploration/breast/base
+  - _self_
+
+original_metadata: "mlflow-artifacts:/61/59fec5d041064ed1b0a7552b663f19bf/artifacts/slides_with_folds.csv"
+
+metadata:
+  run_name: B20-24 Data
+  description: B20-24 Data
diff --git a/configs/exploration/breast/base.yaml b/configs/exploration/breast/base.yaml
new file mode 100644
index 00000000..5ff30a82
--- /dev/null
+++ b/configs/exploration/breast/base.yaml
@@ -0,0 +1,6 @@
+# @package _global_
+
+defaults:
+  - /default
+  - /data/breast/base@_global_
+  - _self_
diff --git a/exploration/b20_24.py b/exploration/b20_24.py
new file mode 100644
index 00000000..f1b23174
--- /dev/null
+++ b/exploration/b20_24.py
@@ -0,0 +1,49 @@
+""" B20-24 Metadata Adapter. """
+
+from pathlib import Path
+from tempfile import TemporaryDirectory
+
+import hydra
+import mlflow
+import pandas as pd
+from omegaconf import DictConfig
+from rationai.mlkit import with_cli_args
+from rationai.mlkit.autolog import autolog
+from rationai.mlkit.lightning.loggers import MLFlowLogger
+
+
+@hydra.main(
+    config_path="../configs",
+    config_name="exploration/breast/b20_24",
+    version_base=None,
+)
+@autolog
+def main(config: DictConfig, logger: MLFlowLogger) -> None:
+    df = pd.read_csv(mlflow.artifacts.download_artifacts(config.original_metadata))
+    df["carcinoma"] = df["carcinoma"].map(lambda x: x == "T") # binarize carcinoma
+    df["slide_path"] = df["path"] # rename path column (unified with prostate)
+    df = df.drop(
+        ["id", "path", "extent_x", "extent_y", "tile_extent_x", "tile_extent_y", "stride_x", "stride_y", "mpp_x", "mpp_y", "level", "slide_id", "total", "pos", "neg"],
+        axis=1
+    ) # remove unnecessary attributes
+
+    val_df = df[ df["split"] == 1 ] # use one fold as validation set (unified with prostate)
+    train_df = df[ df["split"] != 1 ] # use rest as training set
+
+    with TemporaryDirectory() as tmp_dir:
+        target = Path(tmp_dir) / f"b20_24_metadata.csv"
+        df.to_csv(str(target), index=False)
+
+        train_target = Path(tmp_dir) / f"b20_24_train_metadata.csv"
+        train_df.to_csv(str(train_target), index=False)
+
+        val_target = Path(tmp_dir) / f"b20_24_val_metadata.csv"
+        val_df.to_csv(str(val_target), index=False)
+
+        mlflow.log_artifact(str(target))
+        mlflow.log_artifact(str(train_target))
+        mlflow.log_artifact(str(val_target))
+
+
+if __name__ == "__main__":
+    main()

From ccbf8dafccd0ea1feaefa48a649e9cae14f6134d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 10:55:47 +0000
Subject: [PATCH 3/8] feat: exploration

---
 configs/data/breast/mmci/b20_24_base.yaml  |  6 +-
 configs/data/breast/mmci/b20_24_full.yaml  |  9 +++
 configs/data/breast/mmci/b20_24_train.yaml |  4 +-
 configs/data/breast/mmci/b20_24_val.yaml   |  4 +-
 configs/exploration/breast/b20_24.yaml     |  6 +-
 exploration/b20_24.py                      | 84 +++++++++++++++++++---
 6 files changed, 95 insertions(+), 18 deletions(-)
 create mode 100644 configs/data/breast/mmci/b20_24_full.yaml

diff --git a/configs/data/breast/mmci/b20_24_base.yaml b/configs/data/breast/mmci/b20_24_base.yaml
index 641bfb37..dc029cf5 100644
--- a/configs/data/breast/mmci/b20_24_base.yaml
+++ b/configs/data/breast/mmci/b20_24_base.yaml
@@ -3,7 +3,7 @@ defaults:
   - _self_
 
 
-qc_masks_uri: "mlflow-artifacts:/61/1fa3adfff18b433782163a29bc9d5c81/artifacts/qc_masks"
+qc_masks_uri: "mlflow-artifacts:/61/1fa3adfff18b433782163a29bc9d5c81/artifacts/qc_masks_structured"
 tissue_masks_uri: "mlflow-artifacts:/61/b80d4357ba3b44999896ceda971a1884/artifacts/tissue_masks"
 epithelium_masks_uri: "mlflow-artifacts:/61/0af479b624824d9bae0bc830a5b99f9a/artifacts/epithelium_masks"
 
@@ -20,7 +20,3 @@ thresholds:
 
   epithelium_roi_t: 0.2
 
-metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_metadata.csv"
-
-
-data_name: "MMCI B20-24"
diff --git a/configs/data/breast/mmci/b20_24_full.yaml b/configs/data/breast/mmci/b20_24_full.yaml
new file mode 100644
index 00000000..97d524d8
--- /dev/null
+++ b/configs/data/breast/mmci/b20_24_full.yaml
@@ -0,0 +1,9 @@
+defaults:
+  - b20_24_base
+  - _self_
+
+tiles_uri_512: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_512"
+tiles_uri_224: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_224"
+metadata_table: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_metadata.csv"
+
+data_name: "MMCI B20-24"
diff --git a/configs/data/breast/mmci/b20_24_train.yaml b/configs/data/breast/mmci/b20_24_train.yaml
index 81301cd0..32e53fab 100644
--- a/configs/data/breast/mmci/b20_24_train.yaml
+++ b/configs/data/breast/mmci/b20_24_train.yaml
@@ -2,6 +2,8 @@ defaults:
   - b20_24_base
   - _self_
 
-metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_train_metadata.csv"
+tiles_uri_512: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_512"
+tiles_uri_224: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_224"
+metadata_table: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_metadata.csv"
 
 data_name: "MMCI B20-24 Train"
diff --git a/configs/data/breast/mmci/b20_24_val.yaml b/configs/data/breast/mmci/b20_24_val.yaml
index 758cb7c2..175fd5a8 100644
--- a/configs/data/breast/mmci/b20_24_val.yaml
+++ b/configs/data/breast/mmci/b20_24_val.yaml
@@ -2,6 +2,8 @@ defaults:
   - b20_24_base
   - _self_
 
-metadata_table: "mlflow-artifacts:/61/1e5b475e3b2043ba924683d457fdb37b/artifacts/b20_24_val_metadata.csv"
+tiles_uri_512: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_512"
+tiles_uri_224: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_224"
+metadata_table: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_metadata.csv"
 
 data_name: "MMCI B20-24 Val"
diff --git a/configs/exploration/breast/b20_24.yaml b/configs/exploration/breast/b20_24.yaml
index 9bcc2d07..f0a0731f 100644
--- a/configs/exploration/breast/b20_24.yaml
+++ b/configs/exploration/breast/b20_24.yaml
@@ -5,7 +5,9 @@ defaults:
   - _self_
 
 original_metadata: "mlflow-artifacts:/61/59fec5d041064ed1b0a7552b663f19bf/artifacts/slides_with_folds.csv"
+original_tiling_512: "mlflow-artifacts:/61/8a2711c2f25346eea4057a6faa76b048/artifacts/BC_train_data"
+original_tiling_224: "mlflow-artifacts:/61/e77bd9fd006e4618ac8f834411981b2a/artifacts/BC_train_data"
 
 metadata:
-  run_name: B20-24 Data
-  description: B20-24 Data
+  run_name: B20-24 Adapted Data
+  description: B20-24 Adapted Data
diff --git a/exploration/b20_24.py b/exploration/b20_24.py
index f1b23174..c0ec2d42 100644
--- a/exploration/b20_24.py
+++ b/exploration/b20_24.py
@@ -1,4 +1,4 @@
-""" B20-24 Metadata Adapter. """
+"""B20-24 Metadata Adapter."""
 
 from pathlib import Path
 from tempfile import TemporaryDirectory
@@ -7,9 +7,37 @@
 import mlflow
 import pandas as pd
 from omegaconf import DictConfig
-from rationai.mlkit import with_cli_args
 from rationai.mlkit.autolog import autolog
 from rationai.mlkit.lightning.loggers import MLFlowLogger
+from rationai.tiling.writers import save_mlflow_dataset
+
+
+def repair_split_and_log(
+    tiling_path: Path, all_slides: list[str], val_slide_paths: list[str], suffix: str
+) -> None:
+    slides = pd.read_parquet(tiling_path / "slides.parquet")
+    tiles = pd.read_parquet(tiling_path / "tiles.parquet")
+
+    # Binarize carcinoma the same way as the CSV metadata (T -> True, N -> False)
+    slides["carcinoma"] = slides["carcinoma"].map(lambda x: x == "T")
+
+    # Keep only slides (and their tiles) that are actually referenced in the metadata CSV
+    all_slide_mask = slides["path"].isin(all_slides)
+    slides = slides[all_slide_mask]
+    tiles = tiles[tiles["slide_id"].isin(slides["id"])]
+
+    # slides.path matches slide_path from the CSVs
+    val_slide_mask = slides["path"].isin(val_slide_paths)
+    val_slide_ids = slides.loc[val_slide_mask, "id"]
+
+    train_slides, val_slides = slides[~val_slide_mask], slides[val_slide_mask]
+
+    val_tile_mask = tiles["slide_id"].isin(val_slide_ids)
+    train_tiles, val_tiles = tiles[~val_tile_mask], tiles[val_tile_mask]
+
+    save_mlflow_dataset(slides, tiles, f"b20_24_{suffix}")
+    save_mlflow_dataset(train_slides, train_tiles, f"b20_24_train_{suffix}")
+    save_mlflow_dataset(val_slides, val_tiles, f"b20_24_val_{suffix}")
 
 
 @hydra.main(
@@ -20,15 +48,53 @@
 @autolog
 def main(config: DictConfig, logger: MLFlowLogger) -> None:
     df = pd.read_csv(mlflow.artifacts.download_artifacts(config.original_metadata))
-    df["carcinoma"] = df["carcinoma"].map(lambda x: x == "T") # binarize carcinoma
-    df["slide_path"] = df["path"] # rename path column (unified with prostate)
+    original_tiling_512 = Path(
+        mlflow.artifacts.download_artifacts(config.original_tiling_512)
+    )
+    original_tiling_224 = Path(
+        mlflow.artifacts.download_artifacts(config.original_tiling_224)
+    )
+
+    df["carcinoma"] = df["carcinoma"].map(lambda x: x == "T")  # binarize carcinoma
+    df["slide_path"] = df["path"]  # rename path column (unified with prostate)
     df = df.drop(
-        ["id", "path", "extent_x", "extent_y", "tile_extent_x", "tile_extent_y", "stride_x", "stride_y", "mpp_x", "mpp_y", "level", "slide_id", "total", "pos", "neg"],
-        axis=1
-    ) # remove unnecessary attributes
+        [
+            "id",
+            "path",
+            "extent_x",
+            "extent_y",
+            "tile_extent_x",
+            "tile_extent_y",
+            "stride_x",
+            "stride_y",
+            "mpp_x",
+            "mpp_y",
+            "level",
+            "slide_id",
+            "total",
+            "pos",
+            "neg",
+        ],
+        axis=1,
+    )  # remove unnecessary attributes
+
+    val_df = df[
+        df["split"] == 1
+    ]  # use one fold as validation set (unified with prostate)
+    train_df = df[df["split"] != 1]  # use rest as training set
 
-    val_df = df[ df["split"] == 1 ] # use one fold as validation set (unified with prostate)
-    train_df = df[ df["split"] != 1 ] # use rest as training set
+    repair_split_and_log(
+        original_tiling_512,
+        df["slide_path"].tolist(),
+        val_df["slide_path"].tolist(),
+        "512",
+    )
+    repair_split_and_log(
+        original_tiling_224,
+        df["slide_path"].tolist(),
+        val_df["slide_path"].tolist(),
+        "224",
+    )
 
     with TemporaryDirectory() as tmp_dir:
         target = Path(tmp_dir) / f"b20_24_metadata.csv"

From 0d3708e8a23fdd01688be9ba639d8dcafaa9ecd0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 16:34:56 +0000
Subject: [PATCH 4/8] feat: show tile masks

---
 configs/data/breast/mmci/b20_24_train.yaml    |   2 +
 configs/data/breast/mmci/b20_24_val.yaml      |   3 +
 .../filter_tiles/b20_24_train.yaml            |   5 +
 .../filter_tiles/b20_24_val.yaml              |   5 +
 .../tile_masks/b20_24_val_224.yaml            |  15 ++
 .../tile_masks/b20_24_val_512.yaml            |  15 ++
 preprocessing/tiling_v2/filter_tiles.py       |   2 +-
 pyproject.toml                                |   4 +-
 uv.lock                                       | 158 +++++++++---------
 9 files changed, 127 insertions(+), 82 deletions(-)
 create mode 100644 configs/experiment/preprocessing/filter_tiles/b20_24_train.yaml
 create mode 100644 configs/experiment/preprocessing/filter_tiles/b20_24_val.yaml
 create mode 100644 configs/experiment/preprocessing/tile_masks/b20_24_val_224.yaml
 create mode 100644 configs/experiment/preprocessing/tile_masks/b20_24_val_512.yaml

diff --git a/configs/data/breast/mmci/b20_24_train.yaml b/configs/data/breast/mmci/b20_24_train.yaml
index 32e53fab..6b7d7e37 100644
--- a/configs/data/breast/mmci/b20_24_train.yaml
+++ b/configs/data/breast/mmci/b20_24_train.yaml
@@ -4,6 +4,8 @@ defaults:
 
 tiles_uri_512: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_512"
 tiles_uri_224: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_224"
+tiles_filtered_uri_512: "mlflow-artifacts:/61/72b9a94ef0fa49609ea4d19d7b830329/artifacts/MMCI B20-24 Train_512"
+tiles_filtered_uri_224: "mlflow-artifacts:/61/72b9a94ef0fa49609ea4d19d7b830329/artifacts/MMCI B20-24 Train_224"
 metadata_table: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_train_metadata.csv"
 
 data_name: "MMCI B20-24 Train"
diff --git a/configs/data/breast/mmci/b20_24_val.yaml b/configs/data/breast/mmci/b20_24_val.yaml
index 175fd5a8..06725350 100644
--- a/configs/data/breast/mmci/b20_24_val.yaml
+++ b/configs/data/breast/mmci/b20_24_val.yaml
@@ -4,6 +4,9 @@ defaults:
 
 tiles_uri_512: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_512"
 tiles_uri_224: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_224"
+tiles_filtered_uri_512: "mlflow-artifacts:/61/2c4f49ecb47247ca862cdc1c1868d4b2/artifacts/MMCI B20-24 Val_512"
+tiles_filtered_uri_224: "mlflow-artifacts:/61/2c4f49ecb47247ca862cdc1c1868d4b2/artifacts/MMCI B20-24 Val_224"
+
 metadata_table: "mlflow-artifacts:/61/230c412fc727466b888fb4db3625d128/artifacts/b20_24_val_metadata.csv"
 
 data_name: "MMCI B20-24 Val"
diff --git a/configs/experiment/preprocessing/filter_tiles/b20_24_train.yaml b/configs/experiment/preprocessing/filter_tiles/b20_24_train.yaml
new file mode 100644
index 00000000..d4227674
--- /dev/null
+++ b/configs/experiment/preprocessing/filter_tiles/b20_24_train.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/b20_24_train@data
+  - _self_
diff --git a/configs/experiment/preprocessing/filter_tiles/b20_24_val.yaml b/configs/experiment/preprocessing/filter_tiles/b20_24_val.yaml
new file mode 100644
index 00000000..fc6aacbb
--- /dev/null
+++ b/configs/experiment/preprocessing/filter_tiles/b20_24_val.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/b20_24_val@data
+  - _self_
diff --git a/configs/experiment/preprocessing/tile_masks/b20_24_val_224.yaml b/configs/experiment/preprocessing/tile_masks/b20_24_val_224.yaml
new file mode 100644
index 00000000..04f1c172
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_masks/b20_24_val_224.yaml
@@ -0,0 +1,15 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/b20_24_val@data
+  - _self_
+
+tile_uris:
+  - ${data.tiles_filtered_uri_224}
+
+percentage_cols:
+  - "tissue_roi_percentage"
+  - "residual_percentage"
+  - "blur_percentage"
+  - "folding_percentage"
+  - "epithelium_roi_percentage"
diff --git a/configs/experiment/preprocessing/tile_masks/b20_24_val_512.yaml b/configs/experiment/preprocessing/tile_masks/b20_24_val_512.yaml
new file mode 100644
index 00000000..ff35486e
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_masks/b20_24_val_512.yaml
@@ -0,0 +1,15 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/b20_24_val@data
+  - _self_
+
+tile_uris:
+  - ${data.tiles_filtered_uri_512}
+
+percentage_cols:
+  - "tissue_roi_percentage"
+  - "residual_percentage"
+  - "blur_percentage"
+  - "folding_percentage"
+  - "epithelium_roi_percentage"
diff --git a/preprocessing/tiling_v2/filter_tiles.py b/preprocessing/tiling_v2/filter_tiles.py
index 1888444a..7cb54040 100644
--- a/preprocessing/tiling_v2/filter_tiles.py
+++ b/preprocessing/tiling_v2/filter_tiles.py
@@ -13,7 +13,7 @@
 
 def filter_tiles(tiles: pd.DataFrame, thresholds: dict[str, int]) -> pd.DataFrame:
     for col in tiles.columns:
-        if col.endswith("percentage") and "carcinoma" not in col:
+        if col.endswith("percentage") and ("carcinoma" not in col) and ("epithelium" not in col):
             t = col.replace("percentage", "t")
             assert t in thresholds, f"{t} for {col}"
             mask = (
diff --git a/pyproject.toml b/pyproject.toml
index d70316a5..c10b1b55 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,7 +1,7 @@
 [project]
-name = "prostate-cancer"
+name = "carcinoma-binary-classification-methods"
 version = "0.1.0"
-description = "A project for prostate cancer detection in digital pathology slides."
+description = "A project for carcinoma detection in digital pathology slides of various tissues."
 readme = "README.md"
 requires-python = ">=3.12,<3.14"
 license = { file = "LICENSE" }
diff --git a/uv.lock b/uv.lock
index 2836046d..b416e2d4 100644
--- a/uv.lock
+++ b/uv.lock
@@ -216,6 +216,85 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/72/76/20fa66124dbe6be5cafeb312ece67de6b61dd91a0247d1ea13db4ebb33c2/cachetools-5.5.2-py3-none-any.whl", hash = "sha256:d26a22bcc62eb95c3beabd9f1ee5e820d3d2704fe2967cbe350e20c8ffcd3f0a", size = 10080, upload-time = "2025-02-20T21:01:16.647Z" },
 ]
 
+[[package]]
+name = "carcinoma-binary-classification-methods"
+version = "0.1.0"
+source = { virtual = "." }
+dependencies = [
+    { name = "albumentations" },
+    { name = "datasets" },
+    { name = "deprecated" },
+    { name = "huggingface-hub" },
+    { name = "ipykernel" },
+    { name = "lightning" },
+    { name = "matplotlib" },
+    { name = "mlflow" },
+    { name = "numpy" },
+    { name = "openpyxl" },
+    { name = "pandas" },
+    { name = "pillow" },
+    { name = "rationai-masks" },
+    { name = "rationai-mlkit" },
+    { name = "rationai-sdk" },
+    { name = "rationai-staining" },
+    { name = "rationai-tiling" },
+    { name = "ratiopath" },
+    { name = "ray", extra = ["data"] },
+    { name = "regex" },
+    { name = "scikit-learn" },
+    { name = "timm" },
+    { name = "torch" },
+    { name = "torchvision" },
+    { name = "tqdm" },
+    { name = "transformers" },
+    { name = "triton" },
+]
+
+[package.dev-dependencies]
+dev = [
+    { name = "mypy" },
+    { name = "pre-commit" },
+    { name = "ruff" },
+]
+
+[package.metadata]
+requires-dist = [
+    { name = "albumentations", specifier = ">=2.0.8" },
+    { name = "datasets", specifier = ">=4.6.1" },
+    { name = "deprecated", specifier = ">=1.3.1" },
+    { name = "huggingface-hub", specifier = ">=1.12.2" },
+    { name = "ipykernel", specifier = ">=7.2.0" },
+    { name = "lightning", specifier = "==2.6.0" },
+    { name = "matplotlib", specifier = "==3.10.1" },
+    { name = "mlflow", specifier = "==3.12.0" },
+    { name = "numpy", specifier = ">=2.3.2" },
+    { name = "openpyxl", specifier = "==3.1.5" },
+    { name = "pandas", specifier = ">=2.2.3" },
+    { name = "pillow", specifier = "==11.3.0" },
+    { name = "rationai-masks", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/masks.git" },
+    { name = "rationai-mlkit", git = "https://github.com/RationAI/mlkit.git" },
+    { name = "rationai-sdk", git = "https://github.com/RationAI/rationai-sdk-python.git" },
+    { name = "rationai-staining", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/staining.git?rev=2437ec9af319731f3e32aa79c4bf9e6fef5a5711" },
+    { name = "rationai-tiling", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/tiling.git" },
+    { name = "ratiopath", specifier = ">=1.1.2" },
+    { name = "ray", extras = ["data"], specifier = ">=2.52.1" },
+    { name = "regex", specifier = ">=2026.2.28" },
+    { name = "scikit-learn", specifier = "==1.8.0" },
+    { name = "timm", specifier = "==1.0.15" },
+    { name = "torch", specifier = ">=2.9.1" },
+    { name = "torchvision", specifier = ">=0.22.0" },
+    { name = "tqdm", specifier = "==4.67.1" },
+    { name = "transformers", specifier = ">=5.7.0" },
+    { name = "triton", specifier = ">=3.3.0" },
+]
+
+[package.metadata.requires-dev]
+dev = [
+    { name = "mypy" },
+    { name = "pre-commit" },
+    { name = "ruff" },
+]
+
 [[package]]
 name = "certifi"
 version = "2026.4.22"
@@ -2373,85 +2452,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5b/5a/bc7b4a4ef808fa59a816c17b20c4bef6884daebbdf627ff2a161da67da19/propcache-0.4.1-py3-none-any.whl", hash = "sha256:af2a6052aeb6cf17d3e46ee169099044fd8224cbaf75c76a2ef596e8163e2237", size = 13305, upload-time = "2025-10-08T19:49:00.792Z" },
 ]
 
-[[package]]
-name = "prostate-cancer"
-version = "0.1.0"
-source = { virtual = "." }
-dependencies = [
-    { name = "albumentations" },
-    { name = "datasets" },
-    { name = "deprecated" },
-    { name = "huggingface-hub" },
-    { name = "ipykernel" },
-    { name = "lightning" },
-    { name = "matplotlib" },
-    { name = "mlflow" },
-    { name = "numpy" },
-    { name = "openpyxl" },
-    { name = "pandas" },
-    { name = "pillow" },
-    { name = "rationai-masks" },
-    { name = "rationai-mlkit" },
-    { name = "rationai-sdk" },
-    { name = "rationai-staining" },
-    { name = "rationai-tiling" },
-    { name = "ratiopath" },
-    { name = "ray", extra = ["data"] },
-    { name = "regex" },
-    { name = "scikit-learn" },
-    { name = "timm" },
-    { name = "torch" },
-    { name = "torchvision" },
-    { name = "tqdm" },
-    { name = "transformers" },
-    { name = "triton" },
-]
-
-[package.dev-dependencies]
-dev = [
-    { name = "mypy" },
-    { name = "pre-commit" },
-    { name = "ruff" },
-]
-
-[package.metadata]
-requires-dist = [
-    { name = "albumentations", specifier = ">=2.0.8" },
-    { name = "datasets", specifier = ">=4.6.1" },
-    { name = "deprecated", specifier = ">=1.3.1" },
-    { name = "huggingface-hub", specifier = ">=1.12.2" },
-    { name = "ipykernel", specifier = ">=7.2.0" },
-    { name = "lightning", specifier = "==2.6.0" },
-    { name = "matplotlib", specifier = "==3.10.1" },
-    { name = "mlflow", specifier = "==3.12.0" },
-    { name = "numpy", specifier = ">=2.3.2" },
-    { name = "openpyxl", specifier = "==3.1.5" },
-    { name = "pandas", specifier = ">=2.2.3" },
-    { name = "pillow", specifier = "==11.3.0" },
-    { name = "rationai-masks", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/masks.git" },
-    { name = "rationai-mlkit", git = "https://github.com/RationAI/mlkit.git" },
-    { name = "rationai-sdk", git = "https://github.com/RationAI/rationai-sdk-python.git" },
-    { name = "rationai-staining", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/staining.git?rev=2437ec9af319731f3e32aa79c4bf9e6fef5a5711" },
-    { name = "rationai-tiling", git = "https://gitlab.ics.muni.cz/rationai/digital-pathology/libraries/tiling.git" },
-    { name = "ratiopath", specifier = ">=1.1.2" },
-    { name = "ray", extras = ["data"], specifier = ">=2.52.1" },
-    { name = "regex", specifier = ">=2026.2.28" },
-    { name = "scikit-learn", specifier = "==1.8.0" },
-    { name = "timm", specifier = "==1.0.15" },
-    { name = "torch", specifier = ">=2.9.1" },
-    { name = "torchvision", specifier = ">=0.22.0" },
-    { name = "tqdm", specifier = "==4.67.1" },
-    { name = "transformers", specifier = ">=5.7.0" },
-    { name = "triton", specifier = ">=3.3.0" },
-]
-
-[package.metadata.requires-dev]
-dev = [
-    { name = "mypy" },
-    { name = "pre-commit" },
-    { name = "ruff" },
-]
-
 [[package]]
 name = "protobuf"
 version = "5.29.6"

From 1a208295ab4f8014f78b0584fb93d34e3b9b9271 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 17:22:45 +0000
Subject: [PATCH 5/8] feat: mammaprint tile masks

---
 configs/data/breast/mmci/mammaprint_test.yaml | 28 ++++++++++
 .../filter_tiles/mammaprint_test.yaml         |  5 ++
 .../tile_masks/mammaprint_test_512.yaml       | 16 ++++++
 .../exploration/breast/mammaprint_test.yaml   | 13 +++++
 exploration/b20_24.py                         |  2 +-
 exploration/mammaprint_test.py                | 55 +++++++++++++++++++
 6 files changed, 118 insertions(+), 1 deletion(-)
 create mode 100644 configs/data/breast/mmci/mammaprint_test.yaml
 create mode 100644 configs/experiment/preprocessing/filter_tiles/mammaprint_test.yaml
 create mode 100644 configs/experiment/preprocessing/tile_masks/mammaprint_test_512.yaml
 create mode 100644 configs/exploration/breast/mammaprint_test.yaml
 create mode 100644 exploration/mammaprint_test.py

diff --git a/configs/data/breast/mmci/mammaprint_test.yaml b/configs/data/breast/mmci/mammaprint_test.yaml
new file mode 100644
index 00000000..4c2844d8
--- /dev/null
+++ b/configs/data/breast/mmci/mammaprint_test.yaml
@@ -0,0 +1,28 @@
+defaults:
+  - base
+  - _self_
+
+qc_masks_uri: "mlflow-artifacts:/61/a382d706e2c24906a6dbcc070d6ae718/artifacts/qc_masks"
+tissue_masks_uri: "mlflow-artifacts:/61/d5c1eceb2a32492ca605f33a7a13c2f0/artifacts/tissue_masks"
+carcinoma_masks_uri: "mlflow-artifacts:/61/f8a6422e954d4e86b166334a3a1f75ea/artifacts/annotation_masks/carcinoma"
+exclude_masks_uri: "mlflow-artifacts:/61/f8a6422e954d4e86b166334a3a1f75ea/artifacts/annotation_masks/exclude"
+
+# were used for filtering
+thresholds:
+  tissue_roi_t: 0.01
+
+  residual_t: 0.25
+  blur_t: 0.4
+  folding_t: 0.25
+
+  carcinoma_roi_t: 0.8
+  exclude_t: 0.5
+
+
+tiles_uri_512: "mlflow-artifacts:/61/605e6e530d3e48ad8c8d784627b2b1d2/artifacts/mammaprint_test_512"
+tiles_uri_224: "mlflow-artifacts:/61/605e6e530d3e48ad8c8d784627b2b1d2/artifacts/mammaprint_test_224"
+tiles_filtered_uri_512: "mlflow-artifacts:/61/dae5ddebc8234a449ea6ca4c42b6e927/artifacts/Mammaprint Test_512"
+tiles_filtered_uri_224: "mlflow-artifacts:/61/dae5ddebc8234a449ea6ca4c42b6e927/artifacts/Mammaprint Test_224"
+metadata_table: "mlflow-artifacts:/61/605e6e530d3e48ad8c8d784627b2b1d2/artifacts/mammaprint_test_metadata.csv"
+
+data_name: "Mammaprint Test"
diff --git a/configs/experiment/preprocessing/filter_tiles/mammaprint_test.yaml b/configs/experiment/preprocessing/filter_tiles/mammaprint_test.yaml
new file mode 100644
index 00000000..da211984
--- /dev/null
+++ b/configs/experiment/preprocessing/filter_tiles/mammaprint_test.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/mammaprint_test@data
+  - _self_
diff --git a/configs/experiment/preprocessing/tile_masks/mammaprint_test_512.yaml b/configs/experiment/preprocessing/tile_masks/mammaprint_test_512.yaml
new file mode 100644
index 00000000..f62ae761
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_masks/mammaprint_test_512.yaml
@@ -0,0 +1,16 @@
+# @package _global_
+
+defaults:
+  - /data/breast/mmci/mammaprint_test@data
+  - _self_
+
+tile_uris:
+  - ${data.tiles_filtered_uri_512}
+
+percentage_cols:
+  - "tissue_roi_percentage"
+  - "residual_percentage"
+  - "blur_percentage"
+  - "folding_percentage"
+  - "carcinoma_roi_percentage"
+  - "exclude_percentage"
diff --git a/configs/exploration/breast/mammaprint_test.yaml b/configs/exploration/breast/mammaprint_test.yaml
new file mode 100644
index 00000000..107ae987
--- /dev/null
+++ b/configs/exploration/breast/mammaprint_test.yaml
@@ -0,0 +1,13 @@
+# @package _global_
+
+defaults:
+  - /exploration/breast/base
+  - _self_
+
+original_metadata: "mlflow-artifacts:/61/cbccb8df978d4fd7bf896527274fcacc/artifacts/dataframe/dataframe_mammaprint.csv"
+original_tiling_512: "mlflow-artifacts:/61/dada21791ac443d9abfca5c06e603669/artifacts/BC_test_data"
+original_tiling_224: "mlflow-artifacts:/61/468b5864f5824e1486147e1bee48e424/artifacts/BC_test_data"
+
+metadata:
+  run_name: Mammaprint Test Adapted Data
+  description: Mammaprint Test Adapted Data
diff --git a/exploration/b20_24.py b/exploration/b20_24.py
index c0ec2d42..c0cb9f94 100644
--- a/exploration/b20_24.py
+++ b/exploration/b20_24.py
@@ -1,4 +1,4 @@
-"""B20-24 Metadata Adapter."""
+"""B20-24 Data Adapter."""
 
 from pathlib import Path
 from tempfile import TemporaryDirectory
diff --git a/exploration/mammaprint_test.py b/exploration/mammaprint_test.py
new file mode 100644
index 00000000..e4f7bf41
--- /dev/null
+++ b/exploration/mammaprint_test.py
@@ -0,0 +1,55 @@
+""" Mammaprint Test Data Adapter."""
+
+from pathlib import Path
+from tempfile import TemporaryDirectory
+
+import hydra
+import mlflow
+import pandas as pd
+from omegaconf import DictConfig
+from rationai.mlkit.autolog import autolog
+from rationai.mlkit.lightning.loggers import MLFlowLogger
+from rationai.tiling.writers import save_mlflow_dataset
+
+
+def repair_and_log(tiling_path: Path, suffix: str) -> None:
+    slides = pd.read_parquet(tiling_path / "slides.parquet")
+    tiles = pd.read_parquet(tiling_path / "tiles.parquet")
+    slides["carcinoma"] = True # all slides are positive
+    save_mlflow_dataset(slides, tiles, f"mammaprint_test_{suffix}")
+
+
+@hydra.main(
+    config_path="../configs",
+    config_name="exploration/breast/mammaprint_test",
+    version_base=None,
+)
+@autolog
+def main(config: DictConfig, logger: MLFlowLogger) -> None:
+    df = pd.read_csv(mlflow.artifacts.download_artifacts(config.original_metadata))
+    original_tiling_512 = Path(
+        mlflow.artifacts.download_artifacts(config.original_tiling_512)
+    )
+    original_tiling_224 = Path(
+        mlflow.artifacts.download_artifacts(config.original_tiling_224)
+    )
+
+    df["carcinoma"] = True # all slides are positive
+
+    repair_and_log(
+        original_tiling_512,
+        "512",
+    )
+    repair_and_log(
+        original_tiling_224,
+        "224"
+    )
+
+    with TemporaryDirectory() as tmp_dir:
+        target = Path(tmp_dir) / f"mammaprint_test_metadata.csv"
+        df.to_csv(str(target), index=False)
+        mlflow.log_artifact(str(target))
+
+
+if __name__ == "__main__":
+    main()

From fc49eb3da26c62fcd87563a506947a8673e6918e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Mon, 3 Aug 2026 17:34:26 +0000
Subject: [PATCH 6/8] feat: tile embeddings scripts

---
 .../preprocessing/tile_embeddings_v2/pgp/b20_24_train.yaml   | 5 +++++
 .../preprocessing/tile_embeddings_v2/pgp/b20_24_val.yaml     | 5 +++++
 .../tile_embeddings_v2/pgp/mammaprint_test.yaml              | 5 +++++
 .../tile_embeddings_v2/virchow2/b20_24_train.yaml            | 5 +++++
 .../tile_embeddings_v2/virchow2/b20_24_val.yaml              | 5 +++++
 .../tile_embeddings_v2/virchow2/mammaprint_test.yaml         | 5 +++++
 6 files changed, 30 insertions(+)
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_train.yaml
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_val.yaml
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/pgp/mammaprint_test.yaml
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_train.yaml
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_val.yaml
 create mode 100644 configs/experiment/preprocessing/tile_embeddings_v2/virchow2/mammaprint_test.yaml

diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_train.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_train.yaml
new file mode 100644
index 00000000..704da147
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_train.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/pgp/base
+  - /data/breast/mmci/b20_24_train@data
diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_val.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_val.yaml
new file mode 100644
index 00000000..5d7cf4e9
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/b20_24_val.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/pgp/base
+  - /data/breast/mmci/b20_24_val@data
diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/pgp/mammaprint_test.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/mammaprint_test.yaml
new file mode 100644
index 00000000..d02ebd6b
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/pgp/mammaprint_test.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/pgp/base
+  - /data/breast/mmci/mammaprint_test@data
diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_train.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_train.yaml
new file mode 100644
index 00000000..9818526d
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_train.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/virchow2/base
+  - /data/breast/mmci/b20_24_train@data
diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_val.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_val.yaml
new file mode 100644
index 00000000..067688bc
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/b20_24_val.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/virchow2/base
+  - /data/breast/mmci/b20_24_val@data
diff --git a/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/mammaprint_test.yaml b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/mammaprint_test.yaml
new file mode 100644
index 00000000..8b688cfb
--- /dev/null
+++ b/configs/experiment/preprocessing/tile_embeddings_v2/virchow2/mammaprint_test.yaml
@@ -0,0 +1,5 @@
+# @package _global_
+
+defaults:
+  - /experiment/preprocessing/tile_embeddings_v2/virchow2/base
+  - /data/breast/mmci/mammaprint_test@data

From 6502f12511f0772f92a3aeecfb76208cc4a7abfb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Tue, 4 Aug 2026 06:22:56 +0000
Subject: [PATCH 7/8] feat: bigger files

---
 configs/preprocessing/tile_embeddings_v2.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/preprocessing/tile_embeddings_v2.yaml b/configs/preprocessing/tile_embeddings_v2.yaml
index fd978d1f..561c398f 100644
--- a/configs/preprocessing/tile_embeddings_v2.yaml
+++ b/configs/preprocessing/tile_embeddings_v2.yaml
@@ -5,7 +5,7 @@ encoder: ???
 output_path: ${carcinoma_artifacts}/tiles_with_embeddings/${encoder}/${data.data_name}
 concurrency: 64
 block_size: 128
-rows_per_file: 50000
+rows_per_file: 150000
 
 use_case: Tile Embeddings (V2)
 

From 358dec3ad48864d0c71863596c108e46216441d6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Adam=20D=C5=BEado=C5=88?= <524839@mail.muni.cz>
Date: Tue, 4 Aug 2026 18:21:19 +0000
Subject: [PATCH 8/8] feat: breast data init

---
 exploration/b20_24.py                   |  6 +++---
 exploration/mammaprint_test.py          | 13 +++++--------
 preprocessing/tiling_v2/filter_tiles.py |  6 +++++-
 3 files changed, 13 insertions(+), 12 deletions(-)

diff --git a/exploration/b20_24.py b/exploration/b20_24.py
index c0cb9f94..b99c598c 100644
--- a/exploration/b20_24.py
+++ b/exploration/b20_24.py
@@ -97,13 +97,13 @@ def main(config: DictConfig, logger: MLFlowLogger) -> None:
     )
 
     with TemporaryDirectory() as tmp_dir:
-        target = Path(tmp_dir) / f"b20_24_metadata.csv"
+        target = Path(tmp_dir) / "b20_24_metadata.csv"
         df.to_csv(str(target), index=False)
 
-        train_target = Path(tmp_dir) / f"b20_24_train_metadata.csv"
+        train_target = Path(tmp_dir) / "b20_24_train_metadata.csv"
         train_df.to_csv(str(train_target), index=False)
 
-        val_target = Path(tmp_dir) / f"b20_24_val_metadata.csv"
+        val_target = Path(tmp_dir) / "b20_24_val_metadata.csv"
         val_df.to_csv(str(val_target), index=False)
 
         mlflow.log_artifact(str(target))
diff --git a/exploration/mammaprint_test.py b/exploration/mammaprint_test.py
index e4f7bf41..1cba00c9 100644
--- a/exploration/mammaprint_test.py
+++ b/exploration/mammaprint_test.py
@@ -1,4 +1,4 @@
-""" Mammaprint Test Data Adapter."""
+"""Mammaprint Test Data Adapter."""
 
 from pathlib import Path
 from tempfile import TemporaryDirectory
@@ -15,7 +15,7 @@
 def repair_and_log(tiling_path: Path, suffix: str) -> None:
     slides = pd.read_parquet(tiling_path / "slides.parquet")
     tiles = pd.read_parquet(tiling_path / "tiles.parquet")
-    slides["carcinoma"] = True # all slides are positive
+    slides["carcinoma"] = True  # all slides are positive
     save_mlflow_dataset(slides, tiles, f"mammaprint_test_{suffix}")
 
 
@@ -34,19 +34,16 @@ def main(config: DictConfig, logger: MLFlowLogger) -> None:
         mlflow.artifacts.download_artifacts(config.original_tiling_224)
     )
 
-    df["carcinoma"] = True # all slides are positive
+    df["carcinoma"] = True  # all slides are positive
 
     repair_and_log(
         original_tiling_512,
         "512",
     )
-    repair_and_log(
-        original_tiling_224,
-        "224"
-    )
+    repair_and_log(original_tiling_224, "224")
 
     with TemporaryDirectory() as tmp_dir:
-        target = Path(tmp_dir) / f"mammaprint_test_metadata.csv"
+        target = Path(tmp_dir) / "mammaprint_test_metadata.csv"
         df.to_csv(str(target), index=False)
         mlflow.log_artifact(str(target))
 
diff --git a/preprocessing/tiling_v2/filter_tiles.py b/preprocessing/tiling_v2/filter_tiles.py
index 7cb54040..dcbbd518 100644
--- a/preprocessing/tiling_v2/filter_tiles.py
+++ b/preprocessing/tiling_v2/filter_tiles.py
@@ -13,7 +13,11 @@
 
 def filter_tiles(tiles: pd.DataFrame, thresholds: dict[str, int]) -> pd.DataFrame:
     for col in tiles.columns:
-        if col.endswith("percentage") and ("carcinoma" not in col) and ("epithelium" not in col):
+        if (
+            col.endswith("percentage")
+            and ("carcinoma" not in col)
+            and ("epithelium" not in col)
+        ):
             t = col.replace("percentage", "t")
             assert t in thresholds, f"{t} for {col}"
             mask = (