Hugging Face's logo

Datasets:
MedOtter
/
ISPY1-Tumor-SEG-Radiomics

Tasks:
Image Segmentation
Modalities:
Image
Text
3D
Formats:
parquet
Size:
< 1K
Tags:
medical
breast
mri
dce-mri
tumor-segmentation
tcia
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
patient_id
string
image
image
mask
image
overlay
image
slice_index
int32
num_slices
int32
shape
string
spacing_mm
string
mask_fg_voxels_3d
int32
mask_source
string
ISPY1_1002
77
120
(200, 200, 120)
(1.0, 1.0, 1.0)
5,091
masks_stv_manual
ISPY1_1003
63
120
(200, 200, 120)
(1.0, 1.0, 1.0)
13,441
masks_stv_manual
ISPY1_1004
45
90
(180, 180, 90)
(1.0, 1.0, 1.0)
50,679
masks_stv_manual
ISPY1_1005
36
108
(180, 180, 108)
(1.0, 1.0, 1.0)
42,498
masks_stv_manual
ISPY1_1007
72
144
(180, 180, 144)
(1.0, 1.0, 1.0)
9,098
masks_stv_manual
ISPY1_1008
42
112
(200, 200, 112)
(1.0, 1.0, 1.0)
3,920
masks_stv_manual
ISPY1_1009
53
120
(200, 200, 120)
(1.0, 1.0, 1.0)
28,365
masks_stv_manual
ISPY1_1010
37
120
(220, 220, 120)
(1.0, 1.0, 1.0)
71,570
masks_stv_manual
ISPY1_1011
76
120
(210, 210, 120)
(1.0, 1.0, 1.0)
2,472
masks_stv_manual
ISPY1_1012
67
126
(220, 220, 126)
(1.0, 1.0, 1.0)
55,048
masks_stv_manual
ISPY1_1013
53
120
(200, 200, 120)
(1.0, 1.0, 1.0)
2,967
masks_stv_manual
ISPY1_1015
55
120
(200, 200, 120)
(1.0, 1.0, 1.0)
482
masks_stv_manual
ISPY1_1016
46
90
(180, 180, 90)
(1.0, 1.0, 1.0)
12,685
masks_stv_manual
ISPY1_1017
55
120
(200, 200, 120)
(1.0, 1.0, 1.0)
14,379
masks_stv_manual
ISPY1_1021
45
120
(200, 200, 120)
(1.0, 1.0, 1.0)
48,749
masks_stv_manual
ISPY1_1025
57
138
(200, 200, 138)
(1.0, 1.0, 1.0)
11,444
masks_stv_manual
ISPY1_1029
101
138
(200, 200, 138)
(1.0, 1.0, 1.0)
6,227
masks_stv_manual
ISPY1_1030
65
120
(200, 200, 120)
(1.0, 1.0, 1.0)
9,405
masks_stv_manual
ISPY1_1031
99
120
(200, 200, 120)
(1.0, 1.0, 1.0)
28,169
masks_stv_manual
ISPY1_1032
94
120
(200, 200, 120)
(1.0, 1.0, 1.0)
39,372
masks_stv_manual
ISPY1_1034
35
90
(180, 180, 90)
(1.0, 1.0, 1.0)
53,314
masks_stv_manual
ISPY1_1035
43
120
(200, 200, 120)
(1.0, 1.0, 1.0)
118,776
masks_stv_manual
ISPY1_1036
107
150
(200, 200, 150)
(1.0, 1.0, 1.0)
19,656
masks_stv_manual
ISPY1_1037
40
140
(220, 220, 140)
(1.0, 1.0, 1.0)
121,326
masks_stv_manual
ISPY1_1038
95
150
(220, 220, 150)
(1.0, 1.0, 1.0)
14,072
masks_stv_manual
ISPY1_1041
40
138
(200, 200, 138)
(1.0, 1.0, 1.0)
99,233
masks_stv_manual
ISPY1_1042
62
128
(180, 180, 128)
(1.0, 1.0, 1.0)
2,764
masks_stv_manual
ISPY1_1049
52
160
(180, 180, 160)
(1.0, 1.0, 1.0)
486,165
masks_stv_manual
ISPY1_1051
73
120
(200, 200, 120)
(1.0, 1.0, 1.0)
38,398
masks_stv_manual
ISPY1_1053
85
132
(200, 200, 132)
(1.0, 1.0, 1.0)
1,578
masks_stv_manual
ISPY1_1056
48
132
(200, 200, 132)
(1.0, 1.0, 1.0)
72,491
masks_stv_manual
ISPY1_1058
70
140
(180, 180, 140)
(1.0, 1.0, 1.0)
6,448
masks_stv_manual
ISPY1_1060
45
105
(180, 180, 105)
(1.0, 1.0, 1.0)
42,286
masks_stv_manual
ISPY1_1061
75
120
(210, 210, 120)
(1.0, 1.0, 1.0)
22,047
masks_stv_manual
ISPY1_1062
79
120
(200, 200, 120)
(1.0, 1.0, 1.0)
27,652
masks_stv_manual
ISPY1_1064
53
132
(200, 200, 132)
(1.0, 1.0, 1.0)
43,946
masks_stv_manual
ISPY1_1065
22
83
(180, 180, 83)
(1.0, 1.0, 1.0)
24,481
masks_stv_manual
ISPY1_1066
30
150
(180, 180, 150)
(1.0, 1.0, 1.0)
8,603
masks_stv_manual
ISPY1_1069
67
120
(200, 200, 120)
(1.0, 1.0, 1.0)
9,615
masks_stv_manual
ISPY1_1070
68
120
(200, 200, 120)
(1.0, 1.0, 1.0)
7,312
masks_stv_manual
ISPY1_1072
47
132
(200, 200, 132)
(1.0, 1.0, 1.0)
7,228
masks_stv_manual
ISPY1_1073
48
140
(180, 180, 140)
(1.0, 1.0, 1.0)
85,986
masks_stv_manual
ISPY1_1074
41
128
(180, 180, 128)
(1.0, 1.0, 1.0)
1,861
masks_stv_manual
ISPY1_1075
54
120
(210, 210, 120)
(1.0, 1.0, 1.0)
18,438
masks_stv_manual
ISPY1_1077
29
132
(220, 220, 132)
(1.0, 1.0, 1.0)
5,646
masks_stv_manual
ISPY1_1078
59
126
(210, 210, 126)
(1.0, 1.0, 1.0)
53,741
masks_stv_manual
ISPY1_1082
83
132
(200, 200, 132)
(1.0, 1.0, 1.0)
5,446
masks_stv_manual
ISPY1_1083
62
140
(180, 180, 140)
(1.0, 1.0, 1.0)
55,317
masks_stv_manual
ISPY1_1084
49
162
(200, 200, 162)
(1.0, 1.0, 1.0)
60,290
masks_stv_manual
ISPY1_1087
113
155
(180, 180, 155)
(1.0, 1.0, 1.0)
7,375
masks_stv_manual
ISPY1_1088
67
120
(200, 200, 120)
(1.0, 1.0, 1.0)
12,949
masks_stv_manual
ISPY1_1089
76
120
(200, 200, 120)
(1.0, 1.0, 1.0)
142,339
masks_stv_manual
ISPY1_1090
42
132
(200, 200, 132)
(1.0, 1.0, 1.0)
12,594
masks_stv_manual
ISPY1_1094
80
120
(200, 200, 120)
(1.0, 1.0, 1.0)
23,267
masks_stv_manual
ISPY1_1096
61
90
(180, 180, 90)
(1.0, 1.0, 1.0)
7,414
masks_stv_manual
ISPY1_1097
67
147
(200, 200, 147)
(1.0, 1.0, 1.0)
28,810
masks_stv_manual
ISPY1_1098
43
120
(200, 200, 120)
(1.0, 1.0, 1.0)
16,445
masks_stv_manual
ISPY1_1099
49
120
(200, 200, 120)
(1.0, 1.0, 1.0)
36,861
masks_stv_manual
ISPY1_1100
47
120
(200, 200, 120)
(1.0, 1.0, 1.0)
5,834
masks_stv_manual
ISPY1_1101
111
126
(200, 200, 126)
(1.0, 1.0, 1.0)
57,203
masks_stv_manual
ISPY1_1102
127
185
(200, 200, 185)
(1.0, 1.0, 1.0)
9,289
masks_stv_manual
ISPY1_1104
82
120
(220, 220, 120)
(1.0, 1.0, 1.0)
37,480
masks_stv_manual
ISPY1_1106
50
120
(200, 200, 120)
(1.0, 1.0, 1.0)
35,008
masks_stv_manual
ISPY1_1107
85
145
(180, 180, 145)
(1.0, 1.0, 1.0)
3,452
masks_stv_manual
ISPY1_1112
40
140
(180, 180, 140)
(1.0, 1.0, 1.0)
6,496
masks_stv_manual
ISPY1_1114
87
132
(200, 200, 132)
(1.0, 1.0, 1.0)
50,087
masks_stv_manual
ISPY1_1117
68
156
(220, 220, 156)
(1.0, 1.0, 1.0)
5,532
masks_stv_manual
ISPY1_1120
50
120
(180, 180, 120)
(1.0, 1.0, 1.0)
2,639
masks_stv_manual
ISPY1_1122
63
160
(180, 180, 160)
(1.0, 1.0, 1.0)
8,288
masks_stv_manual
ISPY1_1123
49
140
(180, 180, 140)
(1.0, 1.0, 1.0)
45,328
masks_stv_manual
ISPY1_1124
74
120
(200, 200, 120)
(1.0, 1.0, 1.0)
50,219
masks_stv_manual
ISPY1_1125
33
120
(200, 200, 120)
(1.0, 1.0, 1.0)
17,195
masks_stv_manual
ISPY1_1127
59
120
(200, 200, 120)
(1.0, 1.0, 1.0)
4,624
masks_stv_manual
ISPY1_1128
54
120
(200, 200, 120)
(1.0, 1.0, 1.0)
2,269
masks_stv_manual
ISPY1_1129
45
132
(200, 200, 132)
(1.0, 1.0, 1.0)
113,602
masks_stv_manual
ISPY1_1130
90
145
(180, 180, 145)
(1.0, 1.0, 1.0)
2,988
masks_stv_manual
ISPY1_1132
38
132
(220, 220, 132)
(1.0, 1.0, 1.0)
4,936
masks_stv_manual
ISPY1_1134
49
120
(200, 200, 120)
(1.0, 1.0, 1.0)
11,395
masks_stv_manual
ISPY1_1135
83
120
(200, 200, 120)
(1.0, 1.0, 1.0)
5,710
masks_stv_manual
ISPY1_1136
86
140
(180, 180, 140)
(1.0, 1.0, 1.0)
43,649
masks_stv_manual
ISPY1_1137
90
120
(220, 220, 120)
(1.0, 1.0, 1.0)
5,462
masks_stv_manual
ISPY1_1140
36
156
(220, 220, 156)
(1.0, 1.0, 1.0)
18,739
masks_stv_manual
ISPY1_1141
90
150
(200, 200, 150)
(1.0, 1.0, 1.0)
50,445
masks_stv_manual
ISPY1_1142
45
155
(180, 180, 155)
(1.0, 1.0, 1.0)
22,086
masks_stv_manual
ISPY1_1144
56
165
(200, 200, 165)
(1.0, 1.0, 1.0)
40,663
masks_stv_manual
ISPY1_1145
90
165
(200, 200, 165)
(1.0, 1.0, 1.0)
199,832
masks_stv_manual
ISPY1_1146
71
155
(180, 180, 155)
(1.0, 1.0, 1.0)
6,036
masks_stv_manual
ISPY1_1147
49
120
(200, 200, 120)
(1.0, 1.0, 1.0)
50,034
masks_stv_manual
ISPY1_1148
86
128
(180, 180, 128)
(1.0, 1.0, 1.0)
5,620
masks_stv_manual
ISPY1_1149
48
128
(180, 180, 128)
(1.0, 1.0, 1.0)
3,678
masks_stv_manual
ISPY1_1150
75
128
(200, 200, 128)
(1.0, 1.0, 1.0)
22,577
masks_stv_manual
ISPY1_1151
71
147
(200, 200, 147)
(1.0, 1.0, 1.0)
3,966
masks_stv_manual
ISPY1_1152
94
156
(240, 240, 156)
(1.0, 1.0, 1.0)
1,335
masks_stv_manual
ISPY1_1155
53
128
(180, 180, 128)
(1.0, 1.0, 1.0)
53,044
masks_stv_manual
ISPY1_1156
87
132
(200, 200, 132)
(1.0, 1.0, 1.0)
59,502
masks_stv_manual
ISPY1_1157
45
168
(250, 250, 168)
(1.0, 1.0, 1.0)
11,210
masks_stv_manual
ISPY1_1158
95
162
(280, 280, 162)
(1.0, 1.0, 1.0)
3,354
masks_stv_manual
ISPY1_1159
34
128
(180, 180, 128)
(1.0, 1.0, 1.0)
48,567
masks_stv_manual
ISPY1_1160
302
360
(220, 220, 360)
(1.0, 1.0, 1.0)
16,593
masks_stv_manual
ISPY1_1161
105
150
(180, 180, 150)
(1.0, 1.0, 1.0)
50,589
masks_stv_manual
End of preview. Expand in Data Studio

ISPY1-Tumor-SEG-Radiomics

Expert 3D tumor annotations and radiomic features for the I-SPY1 / ACRIN 6657 breast DCE-MRI trial collection. Faithful re-host of the official TCIA analysis-result package (DOI 10.7937/TCIA.XC7A-QT20, v1 2022-06-01, CC BY 3.0), with an added machine-readable evaluation manifest.

  • 163 patients, locally advanced breast cancer, pre-treatment (baseline) visit only
  • Per patient: 3 DCE phases (pre-contrast 0000, early post 0001 ~2.5 min, late post 0002 ~7.5 min) in two preprocessed variants, plus 3 mask tiers
  • Images are T1-weighted DCE-MRI at 1.5 T, multi-center (ACRIN 6657 trial, 2002-2006)

Repository layout

Path Contents Space
NIfTI-Files/images_bias-corrected_nifti/ DICOM->NIfTI, N4 bias-corrected (int16), 489 vols native scanner grid
NIfTI-Files/images_bias-corrected_resampled_zscored_nifti/ + resampled to 1x1x1 mm, z-scored across the 3 phases (float32), 489 vols 1 mm grid
NIfTI-Files/masks_stv_manual/ expert manual structural tumor volume (GOLD standard), binary 1 mm grid (162/163)
NIfTI-Files/masks_ftv/ functional tumor volume (SER-thresholded enhancing voxels, manually cleaned) native grid (all 163)
NIfTI-Files/masks_stv_resunet/ 3D ResUNet predictions (median CV Dice 0.74 vs manual) - never ground truth 1 mm grid (162/163)
derived/masks_stv_manual_aligned_1mm/ repaired mask for ISPY1_1184 (see caveats) 1 mm grid
train.jsonl evaluation manifest: 162 patients, resampled images <-> STV-manual mask -
excluded.jsonl ISPY1_1237 + machine-readable reason -
ispy1_README.txt, ISPY_DataPaper_features.xlsx, captk_parameter_file_ispy1.csv, *.sums official aux files (radiomics were computed from the resampled images + STV-manual masks) -

Recommended pairing (what train.jsonl encodes)

images_bias-corrected_resampled_zscored_nifti (3 channels/phases) with masks_stv_manual. The STV-manual masks live on each patient's 1 mm resampled grid (affines match exactly); masks are binary {0,1}. STV = entire structural primary lesion including non-enhancing regions, delineated on the first post-contrast image by an expert and approved by a board-certified breast radiologist (single rater). Satellite lesions excluded.

Caveats (read before benchmarking)

  1. ISPY1_1237 is excluded from train.jsonl. Its two STV masks (manual + ResUNet) are stored in a synthetic RAS frame (213x200x150 @ 1 mm, identity direction matrix) that matches neither image grid of this patient. An exhaustive 48-orientation x all-offset cross-correlation search against the world-anchored FTV mask and the DCE enhancement map fails to find any anatomically plausible placement (enhancement contrast ~0 under every candidate vs 1.7-6.0 for aligned patients). The patient's images and FTV mask are intact; the raw STV files are kept for provenance. 162 usable patients remain.
  2. ISPY1_1184's official STV-manual mask ships on the native grid (unlike the other 161, which are on the 1 mm grid). derived/masks_stv_manual_aligned_1mm/ISPY1_1184.nii.gz is the world-space nearest-neighbour resample onto that patient's 1 mm image grid (validation: 99.8% of the FTV falls inside it; enhancement contrast in the aligned range). train.jsonl points to the repaired file for this patient (mask_source column).
  3. masks_ftv are on the native grid for all 163 patients and carry a sloppy NIfTI scale factor (scl_slope makes values 0 / 1.0000001) - binarize with > 0.5, and resample to the 1 mm grid via world coordinates if pairing with the resampled images. FTV is an enhancement-derived functional volume, systematically different from structural extent - do not treat it as boundary ground truth.
  4. Orientation is PSR (per NIfTI affines) for images and aligned masks; per-patient FOV crops mean 58 distinct volume shapes - always read geometry from each file's header.
  5. Z-scored images contain negative values (z-units, not intensities); window/normalize per-volume for display or model input.

Leakage / overlap warnings

  • MAMA-MIA (Synapse syn60868042) includes 171 ISPY1 baseline cases and incorporated ~160 of these very masks as its ISPY1 segmentations. Any model trained on MAMA-MIA has seen most of this dataset - do not treat the two as independent benchmarks.
  • BreastDCEDL (Zenodo 15627233) repackages 172 ISPY1 cases (train split).
  • Subset relation: 163 of the 222-patient parent TCIA ISPY1 collection (DOI 10.7937/K9/TCIA.2016.HdHpgJLK). Patient IDs ISPY1_XXXX are consistent across the parent DICOM PatientID, this dataset, the clinical/outcome spreadsheets, MAMA-MIA and BreastDCEDL - use them for deduplication.
  • Distinct cohorts (no known patient overlap): ISPY2/ACRIN 6698, Breast-MRI-NACT-Pilot, Duke-Breast-Cancer-MRI, TCGA-BRCA (TCGA-Breast-Radiogenomics), QIN-Breast.

Provenance & integrity

Downloaded 2026-08-07 from the official TCIA Aspera faspex package (#699) plus the three direct-download aux files on the TCIA analysis-result page. All 1,467 NIfTI files verified against the shipped MD5 manifest ISPY1-Tumor-SEG-Radiomics.sums (paths in that file are relative to the original package root). Payload re-hosted byte-identical; nothing was renamed, re-encoded or resampled except the single documented repair under derived/.

Loading example 

import json, nibabel as nib
from huggingface_hub import hf_hub_download

manifest = hf_hub_download("MedOtter/ISPY1-Tumor-SEG-Radiomics", "train.jsonl", repo_type="dataset")
rows = [json.loads(l) for l in open(manifest)]
row = rows[0]
img = nib.load(hf_hub_download("MedOtter/ISPY1-Tumor-SEG-Radiomics", row["image"][1], repo_type="dataset"))
msk = nib.load(hf_hub_download("MedOtter/ISPY1-Tumor-SEG-Radiomics", row["mask"], repo_type="dataset"))

A preview config with middle-tumor-slice renderings is provided for the Dataset Viewer.

Citations

Chitalia R, Pati S, Bhalerao M, Thakur SP, Jahani N, Belenky V, McDonald ES, Gibbs J, Newitt DC, Hylton NM, Kontos D, Bakas S. Expert tumor annotations and radiomics for locally advanced breast cancer in DCE-MRI for ACRIN 6657/I-SPY1. Scientific Data 9, 440 (2022). doi:10.1038/s41597-022-01555-4

Chitalia R, et al. (2021). Expert tumor annotations and radiomic features for the ISPY1/ACRIN 6657 trial data collection [Data set]. The Cancer Imaging Archive. doi:10.7937/TCIA.XC7A-QT20

Newitt D, Hylton N, on behalf of the I-SPY 1 Network and ACRIN 6657 Trial Team (2016). Multi-center breast DCE-MRI data and segmentations from patients in the I-SPY 1/ACRIN 6657 trials [Data set]. The Cancer Imaging Archive. doi:10.7937/K9/TCIA.2016.HdHpgJLK

Hylton NM, et al. Locally advanced breast cancer: MR imaging for prediction of response to neoadjuvant chemotherapy - results from ACRIN 6657/I-SPY TRIAL. Radiology 263(3):663-672 (2012). doi:10.1148/radiol.12110748
Downloads last month
39
Number of rows:
162
Total file size:
6.58 GB