Table 4 Metadata fields included in the harmonized CSV file.

From: BreastDCEDL: A standardized deep learning-ready breast DCE-MRI dataset of 2,070 patients

Category

Field

Description

Identifiers

 

pid

Unique patient identifier

 

dataset

Source cohort (spy1, spy2, duke)

Spatial Parameters

 

n_xy, n_z

Image dimensions (pixels, slices)

 

xy_spacing, slice_thick

Pixel spacing, slice thickness (mm)

Temporal Indices

 

n_times

Number of timepoints

 

pre, post_early, post_late

Acquisition indices

Tumor Localization

 

mask_start, mask_end

Tumor slice range (z)

 

sraw, eraw, scol, ecol

Bounding box coordinates

 

tum_vol

Tumor volume (cm3)

Demographics

 

age

Patient age (years)

 

menopause

Menopausal status

 

race_white, race_black

Race indicators

Clinical Biomarkers

 

HR

Hormone receptor status (0/1/NaN)

 

HER2

HER2 status (0/1/NaN)

 

pCR

Pathologic complete response (0/1/NaN)

Data Splits

 

test

Split assignment (0=train, 1=test, 2=val)

  1. Spatial parameters enable image harmonization; clinical variables support predictive modeling. Acquisition-specific parameters (scanner vendor, field strength, TR, TE) can be extracted from original TCIA DICOM files using provided code utilities.