id stringlengths 38 43 | image unknown | inst unknown | type unknown | sem unknown | core_id stringclasses 55
values | patient_id stringclasses 11
values | source_split stringclasses 1
value | x int32 0 1.41k | y int32 0 928 | mpp float64 0.38 0.38 | partial_instance_ids listlengths 0 81 | source_license stringclasses 1
value | source_mask_kind stringclasses 1
value | quality_status stringclasses 2
values | human_training_approved bool 1
class | ignored_regions listlengths 0 0 | skip_training bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
phillips2021immune_bsrmsvtc_0000_x0_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nDT9iZIcSZauiemuZu4RASCzqruGc0VIkXn/Z+EbUMj(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAApU0lEQVR4nO2dDaxcZZ3GX6SFLsgiqEviyiZ0lxhQdxE0jWti2W4(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAeR0lEQVR4nO1d247bOBKVOwGCFdFuYCBN///neSQMEGeo3XlIsvB(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 0 | 0 | 0.37744 | [
8,
18,
47,
167,
223,
233,
234,
237,
242,
243,
248
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x128_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nET9+ZMlWXbfid39ur8llsyspRuNxjocAUOjKHFEiRz(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAA5bklEQVR4nO2dC8xeVZnv1y5fobRgp1Vo5CKnjIboCCJiEPUABwN(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAnhUlEQVR4nO1d247jOA5VpRtojIVKA4t46v8/rybGAJ1eZXYeumu(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 128 | 0 | 0.37744 | [
8,
18,
47,
167,
242,
270,
293,
304,
320,
336,
338,
339,
341,
344,
345
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x256_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nDz96bMt63bWib19Zs65ut2c5ja6kpAACwiMTZkwlMP(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAA9X0lEQVR4nO2dC8wlVZXv9/7sbpvm0QIXidiOQiRGJ47gI4yPES7(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAom0lEQVR4nO1da4/jts72TAdYNMRMgYOk/f8/b5ugQGfBnFPgLfZ(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 256 | 0 | 0.37744 | [
52,
55,
57,
167,
242,
270,
293,
304,
320,
360,
388,
403,
418
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x384_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nDz9ac8ta5rfCd1zRKz1THs4Q2ZlZY1tU27LGNwY7Ea(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAyrElEQVR4nO2dbcxmVX3u13p8mDOMwAgUDDjUjpEQ21iB40ilrZA(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAiBElEQVR4nO1dC4sjOQ6u7m0YLqJ7YUm4///zhoSF6cG5G7hljjw(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 384 | 0 | 0.37744 | [
125,
142,
143,
146,
148,
149,
167,
242,
270,
293,
304,
320,
360,
388,
403,
455,
497,
506
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x512_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nET92bNt2ZXeh81+rb1Pe5u82QGJQrWsYpkqizZtUQ4(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAl70lEQVR4nO2dX6wtVX3HZx0vN3AFroBg9EKV1L74UI0RU1CpiYn(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAYXUlEQVR4nO1d2Y4kqQ6lqlNqTSL109T/f2DNU0vOq5amNFeZlUs(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 512 | 0 | 0.37744 | [
223,
233,
234,
237,
242,
243,
248,
270,
293,
304,
320,
360,
388,
403,
455,
497,
539,
554
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x640_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nEz9Wa9tWXqmh412zrXPiTYzmW2RRVZJKlm2b/wvDfj(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAV5klEQVR4nO3dTah1V3kH8L1P34YUWmNFMjEZOXMoWOg4A6eiCJq(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAO0klEQVR4nO2d22ojOxBFO7Zh4Ajy5v//QL8FNDCQwMFOHPdF90u(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 640 | 0 | 0.37744 | [
336,
338,
339,
341,
344,
345,
360,
388,
403,
455,
497,
539,
554,
614,
631
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x768_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nET9iZYkyXFtiepsHlkAb3f///eRQKEAEENluJlOb+2(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAPu0lEQVR4nO3dP6/k1n0G4BlikVYIEKixValTKcAGUqtQHajTCqk(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAK0ElEQVR4nO3dTXObuhoAYNvJTBda9///wLvWIovM3Glz0jopoG8(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 768 | 0 | 0.37744 | [
418,
455,
497,
539,
554,
614,
687,
718,
726
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x896_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nET92ZIs2ZVtie1ezU8AN4svfKj//6J6pAhFihQhEz2(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAPpklEQVR4nO3dz4okx4EH4OooefAaw+5lLuvLwggMA7roFfQERje(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAKvUlEQVR4nO3dy5KbuhYAUJI+VRlonP//wIw1yCBVt+5JctId89A(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 896 | 0 | 0.37744 | [
497,
506,
539,
554,
614,
687,
736,
759,
780,
794,
807,
813
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x1024_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nET9iZYkx5Vlicqoah5gVv3//9XrJAmALCaJcFMZ39p(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAUcklEQVR4nO3dO6hs12EG4NHsa2PFLgKOwbgKSCC4Qk1Qb5ArN4Y(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAANZ0lEQVR4nO3dyW7cOBCAYdoIYCB19vs/YM41gG8DJ47TalHclyL(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 1,024 | 0 | 0.37744 | [
554,
614,
687,
736,
759,
780,
820,
841,
861,
872,
884,
892,
893
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
phillips2021immune_bsrmsvtc_0000_x1152_y0 | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nDz92ZYkyXVli0qrah4J1vn/f7tv54yqYiIBkAAi3FT(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIAEAAAAACBg1dlAAAYvElEQVR4nO3dPchc153H8Tv3PDGRIohBCWxSLdgQiAkGy33Abuz(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAAAAADRE4smAAAP+klEQVR4nO3dy27jOBBAUdsI0EBq7fn/72uvK4B3g+44iW3xUSS(...TRUNCATED) | null | phillips2021immune_bsrmsvtc_0000 | bsrmsvtc | train | 1,152 | 0 | 0.37744 | [
631,
687,
736,
759,
780,
820,
841,
861,
872,
893,
973,
974,
977,
981,
993
] | CC BY 4.0 | original cell regions | unreviewed | null | [] | false |
Phillips CTCL
H&E images of skin biopsies from people with cutaneous T-cell lymphoma, with cell-region instance masks and cell types from paired CODEX imaging. The source study examined how the arrangement of immune and tumour cells relates to response to pembrolizumab. This dataset contains 69 paired tissue regions from 14 patients.
The data is licensed under CC BY 4.0. Credit Phillips and colleagues and
SPORA, link the license and describe any changes. The original license and
citation are included as LICENSE and reference.bib.
Images and annotations
Images keep their original RGB values at 0.37744 micrometres per pixel. Each crop is 512 × 512 pixels. Crops start 128 pixels apart, with the final row and column placed against the core edge. Each 1,440 × 1,920-pixel core contributes 108 overlapping crops.
The instance masks keep the original cell-region IDs. Use core_id and the
instance ID together to identify a cell across crops.
The 23 cell types retain the source table's new_cluster names. They include
T-cell groups, B cells, plasma cells, macrophages, dendritic cells, Langerhans
cells, mast cells, neutrophils, stromal cells, epithelium and tumour cells.
Here, tumour cells are malignant T cells; they should not be treated as
epithelial tumour cells. classes.json maps the names to mask codes.
Background is zero; 255 is reserved for unknown labels.
There are 107,331 distinct cells in the masks.
The source cell table contains 9,839
additional cell records without an object in the exported masks. These records
are retained with present_in_mask=false in cell-annotations.parquet.
Instance colours distinguish cell regions. Cell-type colours follow the legend. Black is background.
Cell-type instance counts
These counts include only cells present in the crop masks. Each source cell is counted once using its slide and original instance ID, even when crops overlap. Partial cells at crop edges are included. Counts include all quality flags; Unknown and source quality labels remain separate from biological classes.
| Original class | Train | Test | Unique total | Crop appearances |
|---|---|---|---|---|
| B cells | 1,755 | 284 | 2,039 | 26,219 |
| CD4+ T cells | 848 | 278 | 1,126 | 14,800 |
| CD8+ T cells | 3,742 | 947 | 4,689 | 60,536 |
| CLA+ leukocytes | 117 | 54 | 171 | 2,242 |
| DCs, CD11c+ | 30 | 144 | 174 | 2,117 |
| IDO+ stromal cells | 173 | 15 | 188 | 2,500 |
| Langerhans cells | 1,337 | 231 | 1,568 | 19,223 |
| PD1+CD4+ T cells | 271 | 115 | 386 | 4,788 |
| PD1+CD8+ T cells | 148 | 178 | 326 | 3,831 |
| Tregs | 6,863 | 579 | 7,442 | 91,763 |
| epithelium | 9,779 | 5,483 | 15,262 | 189,546 |
| lymphatics | 479 | 197 | 676 | 8,430 |
| macrophages (M1=M2) | 3,266 | 672 | 3,938 | 49,044 |
| macrophages (M1>M2) | 10,432 | 2,137 | 12,569 | 157,305 |
| macrophages (M2>M1) | 843 | 815 | 1,658 | 20,523 |
| mast cells | 479 | 98 | 577 | 7,567 |
| nerves | 418 | 86 | 504 | 5,883 |
| neutrophils | 321 | 74 | 395 | 5,442 |
| plasma cells | 563 | 120 | 683 | 8,670 |
| stroma | 7,344 | 2,510 | 9,854 | 123,904 |
| tumor cells | 25,686 | 11,042 | 36,728 | 465,815 |
| tumor cells, intraepithelial | 652 | 632 | 1,284 | 16,832 |
| vasculature | 4,046 | 1,048 | 5,094 | 66,267 |
| Unknown | 0 | 0 | 0 | 0 |
| Total | 79,592 | 27,739 | 107,331 | 1,353,247 |
instance-counts.csv has one row per crop in train.parquet
or test.parquet. Match id to the crop row; source_file names its Parquet file.
It records total_instances and one class:<original name> column per class,
including zero counts. It also includes the split, slide, patient, crop position,
review status and skip_training flag. Background is not counted.
Within a crop, each instance ID counts once. Summing rows counts a cell again
wherever it appears in overlapping crops; the unique totals above avoid this.
Splits
The original patient splits are retained. Repeat biopsies from the same patient stay together. Do not split overlapping crops into different partitions.
| Split | Crops | Tissue regions | Patients |
|---|---|---|---|
| Train | 5,940 | 55 | 11 |
| Test | 1,512 | 14 | 3 |
| Total | 7,452 | 69 | 14 |
Annotation quality
These are source annotations and may miss cells or have imperfect boundaries.
A visual review of 24 selected training crops found all 24 usable, with good
image-to-mask alignment. These crops have quality_status="usable" and
human_training_approved=true. The remaining 7,428 crops are unreviewed and
have null approval. All test crops remain unreviewed.
No rejected areas were marked, so no crops are flagged for skipping.
skip_training=false does not imply visual approval. Review applies to the
exact inspected crops; overlapping crops are not automatically approved.
The selected review sample does not establish a dataset-wide quality rate
or validate every cell-type label. reviewed-training-index.parquet lists
the 24 reviewed crops; crop-quality.parquet records flags for all crops.
Files and columns
train.parquet and test.parquet contain one row per crop.
| Column | Contents |
|---|---|
id |
Unique crop ID |
image |
PNG bytes; uint8 RGB, height × width × 3 |
inst |
PNG bytes; uint16 cell-region IDs, height × width; background 0 |
type |
PNG bytes; uint8 cell types, height × width; background 0, unknown 255 |
sem |
Null |
core_id, patient_id, source_split |
Source region, patient and split |
x, y, mpp |
Crop origin in pixels and micrometres per pixel |
partial_instance_ids |
IDs of cells that extend outside the crop |
quality_status, human_training_approved |
Review status and nullable approval |
skip_training, ignored_regions |
Recorded rejection flag and crop-local rejected rectangles |
source_license, source_mask_kind |
License and annotation description |
cell-annotations.parquet retains source cell IDs, original and updated class
names, source categories and whether the cell appears in a mask.
crop-instances.parquet lists each cell's crop appearances and pixel counts.
Cells can appear in several crops; count distinct core/instance pairs when
counting cells. training-index.parquet lists the training crop IDs.
Loading and viewing examples
from io import BytesIO
import json
import matplotlib.pyplot as plt
from matplotlib.patches import Patch
import numpy as np
import polars as pl
from PIL import Image
from huggingface_hub import hf_hub_download
from skimage.segmentation import find_boundaries
def read_png(data: bytes) -> np.ndarray:
with Image.open(BytesIO(data)) as image:
return np.array(image)
repo = "csmp-hub/phillips-ctcl"
path = hf_hub_download(repo_id=repo, filename="train.parquet", repo_type="dataset")
classes_path = hf_hub_download(repo_id=repo, filename="classes.json", repo_type="dataset")
with open(classes_path) as stream:
classes = json.load(stream)["classes"]
classes.setdefault("Unknown", 255)
palette = np.vstack((plt.get_cmap("tab20").colors, plt.get_cmap("Set3").colors))
palette = palette[np.ptp(palette, axis=1) > 0.1]
colours = np.zeros((256, 3))
greys = {"unknown": 0.35, "undefined": 0.55, "dirt": 0.7, "excluded": 0.45, "low": 0.8}
for name, code in classes.items():
if code:
colours[code] = greys.get(name.lower(), palette[(code - 1) % len(palette)])
rows = pl.read_parquet(path, n_rows=2)
fig, axes = plt.subplots(2, 4, figsize=(17, 10))
present = set()
for row, panels in zip(rows.iter_rows(named=True), axes):
image = read_png(row["image"])
instances = read_png(row["inst"])
cell_types = read_png(row["type"])
present.update(np.unique(cell_types).tolist())
overlay = image.copy()
overlay[find_boundaries(instances, mode="inner")] = (0, 255, 120)
masks = plt.get_cmap("turbo")((instances.astype(np.uint32) % 251 * 37 % 251) / 250)[..., :3]
masks[instances == 0] = 0
for panel, values, title in zip(
panels,
(image, overlay, masks, colours[cell_types]),
("H&E", "Boundaries", "Instances", "Cell types"),
):
panel.imshow(values)
panel.set_title(title)
panel.axis("off")
handles = [Patch(facecolor=colours[code], label=name)
for name, code in classes.items() if code and code in present]
fig.legend(handles=handles, loc="lower center", ncol=4, fontsize=8)
fig.subplots_adjust(left=0.01, right=0.99, top=0.97, bottom=0.13, wspace=0.04, hspace=0.1)
plt.show()
Sources
- Downloads last month
- 174

