Add he2mif/: ORION-CRC example slice for 5-HE2mIF (600 train / 120 val / 120 test tiles)

README.md

@@ -0,0 +1,81 @@
+---
+license: mit
+pretty_name: MuPaD examples
+tags:
+  - pathology
+  - computational-pathology
+  - virtual-staining
+  - multiplex-immunofluorescence
+  - spatial-transcriptomics
+size_categories:
+  - n<1K
+---
+
+# MuPaD examples
+
+Small, ready-to-run slices of the datasets behind the MuPaD finetuning
+projects in [github.com/jinxixiang/MUPAD](https://github.com/jinxixiang/MUPAD).
+Each folder mirrors the layout of the full dataset, so the released code runs
+on it unchanged: the demos, prediction and scoring, and short training runs.
+
+```
+he2mif/   H&E -> 16-channel mIF      (5-HE2mIF)   ORION-CRC tiles
+he2st/    H&E -> spatial transcriptomics (6-HE2ST)  HEST-1k Xenium slides
+```
+
+Download one folder at a time:
+
+```bash
+huggingface-cli download xiangjx/MuPaD-examples --repo-type dataset --include "he2mif/*" --local-dir examples
+huggingface-cli download xiangjx/MuPaD-examples --repo-type dataset --include "he2st/*"  --local-dir examples
+```
+
+## he2mif/ORIONCRC_dataset_tile_20x
+
+A 393 MB slice of the **ORION-CRC** tile dataset at 20x (512 x 512 px H&E,
+333 x 333 px 17-channel mIF and nuclei masks), in the same layout as the full
+[Zenodo record 15340874](https://zenodo.org/records/15340874):
+
+```
+he/                      H&E tiles (.jpeg)
+if/                      mIF tiles, 17 channels (.tiff, uint8)
+nuclei/                  nuclei instance masks (.tiff, int32 slide-global labels)
+csv_nuclei_pos/          per-slide nuclei positivity tables (.csv)
+train_dataframe.csv      600 tiles, 100 from each of 6 training slides
+val_dataframe.csv        120 tiles,  60 from each of the 2 validation slides
+test_dataframe.csv       120 tiles,  60 from each of the 2 test slides
+demo_dataframe.csv       the 2 validation tiles shown in 5-HE2mIF/demo.ipynb
+slide_dataframe.csv, channel_stats.json
+```
+
+**Selection.** Tiles with tissue (`Hoechst_prop >= 0.2`) were sampled uniformly
+at random per slide (seed 0), so the split is representative of tissue tiles,
+not cherry-picked. The two demo tiles are the exception: one from each
+validation slide, both in the top few percent of the split by patch PCC
+(0.62 and 0.53 against a split median of ~0.32), chosen to make the demo
+legible. Treat the demo as a best case; score the val or test split here for
+representative numbers.
+
+**Positivity tables.** The full per-slide `csv_nuclei_pos/*.csv` run to
+300 MB each. Here each is cut down to the nuclei that appear in the selected
+tiles' masks (every mask label was found), 1-3 MB per slide, which is all the
+stage-1 loss weighting needs.
+
+Point the code at it:
+
+```bash
+export ORION_DATA_ROOT=$PWD/examples/he2mif/ORIONCRC_dataset_tile_20x
+```
+
+Source: the MIPHEI-ViT processed Orion-CRC release, redistributed under its
+MIT license.
+
+> G. Balezo et al. "MIPHEI-ViT: Multiplex Immunofluorescence Prediction from
+> H&E Images using ViT Foundation Models," 2025.
+
+## he2st/
+
+Two processed HEST-1k Xenium slides for `6-HE2ST`: one training slide and the
+held-out validation slide NCBI920, with their raw HEST files and per-cell MUSK
+features. See `6-HE2ST/README.md` for the environment variables that point the
+code at them.

he2mif/ORIONCRC_dataset_tile_20x/channel_stats.json

@@ -0,0 +1,87 @@
+{
+    "Hoechst": {
+        "idx_channel": 0,
+        "std": 38.31878546041518,
+        "is_structural": true
+    },
+    "CD31": {
+        "idx_channel": 1,
+        "std": 5.498713573753559,
+        "is_structural": false
+    },
+    "CD45": {
+        "idx_channel": 2,
+        "std": 26.07032245155099,
+        "is_structural": false
+    },
+    "CD68": {
+        "idx_channel": 3,
+        "std": 10.648181066918182,
+        "is_structural": false
+    },
+    "CD4": {
+        "idx_channel": 4,
+        "std": 15.885813367490442,
+        "is_structural": false
+    },
+    "FOXP3": {
+        "idx_channel": 5,
+        "std": 3.615590655890493,
+        "is_structural": false
+    },
+    "CD8a": {
+        "idx_channel": 6,
+        "std": 8.519019158228375,
+        "is_structural": false
+    },
+    "CD45RO": {
+        "idx_channel": 7,
+        "std": 14.06807888525259,
+        "is_structural": false
+    },
+    "CD20": {
+        "idx_channel": 8,
+        "std": 8.465183851881198,
+        "is_structural": false
+    },
+    "PD-L1": {
+        "idx_channel": 9,
+        "std": 12.57477110852755,
+        "is_structural": false
+    },
+    "CD3e": {
+        "idx_channel": 10,
+        "std": 13.37003412805685,
+        "is_structural": false
+    },
+    "CD163": {
+        "idx_channel": 11,
+        "std": 10.834632444526331,
+        "is_structural": false
+    },
+    "E-cadherin": {
+        "idx_channel": 12,
+        "std": 22.313244729783626,
+        "is_structural": true
+    },
+    "PD-1": {
+        "idx_channel": 13,
+        "std": 20.619796532583972,
+        "is_structural": false
+    },
+    "Ki67": {
+        "idx_channel": 14,
+        "std": 10.465971986050734,
+        "is_structural": false
+    },
+    "Pan-CK": {
+        "idx_channel": 15,
+        "std": 25.021185269910358,
+        "is_structural": true
+    },
+    "SMA": {
+        "idx_channel": 16,
+        "std": 15.16689720673921,
+        "is_structural": true
+    }
+}

he2mif/ORIONCRC_dataset_tile_20x/demo_dataframe.csv

@@ -0,0 +1,3 @@
+in_slide_name,image_path,target_path,nuclei_path,Hoechst_prop,CD31_prop,CD45_prop,CD68_prop,CD4_prop,FOXP3_prop,CD8a_prop,CD45RO_prop,CD20_prop,PD-L1_prop,CD3e_prop,CD163_prop,E-cadherin_prop,PD-1_prop,Ki67_prop,Pan-CK_prop,SMA_prop,CD31_count,CD45_count,CD68_count,CD4_count,FOXP3_count,CD8a_count,CD45RO_count,CD20_count,PD-L1_count,CD3e_count,CD163_count,E-cadherin_count,PD-1_count,Ki67_count,Pan-CK_count,SMA_count
+19510_C19_US_SCAN_OR_001__153041-registered.ome,he/19510_C19_US_SCAN_OR_001__153041-registered.ome_18944_16384_0_512_512.jpeg,if/P37_S51_Full_A24_C59mX_E15_20220129_015300_669681-zlib.ome_18944_16384_0_512_512.tiff,nuclei/P37_S51_Full_A24_C59mX_E15_20220129_015300_669681-zlib.ome_18944_16384_0_512_512.tiff,0.48771667,0.0,0.0398407,0.01802063,0.08734131,0.00033569336,0.000289917,0.00010681152,0.0,0.0,0.0018615723,0.006500244,0.33851624,1.0,0.0101623535,0.38839722,0.5009155,0,6,3,2,0,0,0,0,0,2,0,113,0,8,118,49
+19510_C30_US_SCAN_OR_001__155702-registered.ome,he/19510_C30_US_SCAN_OR_001__155702-registered.ome_45056_3584_0_512_512.jpeg,if/P37_S67_Full_A24_C59nX_E15_20220224_011408_506939-zlib.ome_45056_3584_0_512_512.tiff,nuclei/P37_S67_Full_A24_C59nX_E15_20220224_011408_506939-zlib.ome_45056_3584_0_512_512.tiff,0.8398285,0.0016937256,0.29551697,0.07929993,0.29252625,0.009719849,0.016830444,0.108306885,0.0059051514,0.10978699,0.100753784,0.08256531,0.80241394,1.0,0.6394043,0.7919159,0.009918213,0,70,17,44,8,6,31,2,25,49,20,193,37,214,194,1