C
File size: 3,323 Bytes
1826e37
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b6315dd
1826e37
 
 
 
b6315dd
 
 
 
 
1826e37
b6315dd
1826e37
b6315dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1826e37
 
 
 
 
 
 
 
 
 
 
b6315dd
1826e37
b6315dd
 
 
1826e37
 
 
 
b6315dd
 
 
 
 
 
 
 
 
1826e37
 
b6315dd
1826e37
b6315dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
---
language:
- en
license: cc-by-4.0
task_categories:
- image-classification
tags:
- medical
- cystoscopy
- bladder-cancer
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
---

# CystoDS

Cystoscopic image dataset with lesion annotations, segmentation masks (768 of
8 067 images), and rich clinical metadata (class, subclass, stage, morphology,
modality). Flattened into a single `train` split with one row per image;
images are embedded in the Parquet shards as HF `Image` features and
segmentation polygons are stored inline.

## Why no splits

This dataset ships **one `train` split only**. Use the `pid` (patient id)
column to build your own patient-grouped splits to prevent leakage. 160 unique
patients.

## Columns

| Column | Type | Description |
|---|---|---|
| `image` | `Image` | Decoded PIL image (embedded bytes). |
| `filename` | `string` | 8-char de-identified PNG filename. |
| `pid` | `int64` | Patient id — use this for group-aware splitting. |
| `visit` | `float64` | Visit number (often NaN). |
| `lesion` | `string` | Lesion descriptor (often NaN). |
| `multifocal` | `string` | Multifocal flag (often NaN). |
| `bca` | `string` | Bladder cancer association (`Yes`/`No`). |
| `class` | `string` | Top-level class (see distribution below). |
| `subclass` | `string` | Subclass label. |
| `subclass2` | `string` | Secondary subclass. |
| `stage` | `string` | Tumor stage. |
| `morphology` | `string` | Morphology descriptor. |
| `modality` | `string` | Imaging modality (`WLC`, etc.). |
| `has_segmentation` | `int64` | `1` if a segmentation polygon exists, else `0`. |
| `segmentation` | `struct` | `null` when `has_segmentation == 0`, otherwise `{label: list[string], points: list[list[list[float]]]}` — each `points[i]` is a polygon `[x, y]` list matching `label[i]`. |

## Class distribution

| Class | Count |
|---|---|
| Normal mucosa | 6386 |
| Malignant | 998 |
| Foreign bodies | 251 |
| Non-malignant | 221 |
| Anatomical landmarks | 211 |

## Segmentation labels

768 images have polygons (807 polygon shapes total). Labels include: `Tumor`
(459), `Air bubble` (216), `Right ureteral orifice` (50), `Left ureteral
orifice` (49), `Resection scar` (30), `Flat Tumor` (3).

## Loading

```python
from datasets import load_dataset
ds = load_dataset("milkyroad/C", split="train")
print(ds[0]["image"])          # PIL.Image
print(ds[0]["class"])
# segmentation
if ds[0]["has_segmentation"]:
    seg = ds[0]["segmentation"]
    for label, poly in zip(seg["label"], seg["points"]):
        print(label, len(poly), "points")
```

## Group-aware split example

```python
import random
ds = load_dataset("milkyroad/C", split="train")
pids = sorted({r["pid"] for r in ds if r["pid"] is not None})
random.Random(42).shuffle(pids)
n_test, n_val = 16, 16
test_pids = set(pids[:n_test])
val_pids = set(pids[n_test:n_test + n_val])
train = ds.filter(lambda r: r["pid"] not in test_pids and r["pid"] not in val_pids)
val   = ds.filter(lambda r: r["pid"] in val_pids)
test  = ds.filter(lambda r: r["pid"] in test_pids)
```

## Notes

- 160 unique patients (`pid`). Build splits by **patient** to prevent leakage.
- `segmentation` is `null` for images without a mask (`has_segmentation == 0`).
- Original source: https://osf.io/xvdhy/