ACRIN 6698/I-SPY2 Breast DWI

ACRIN 6698/I-SPY2 Breast DWI is the imaging collection of the ACRIN 6698 trial (NCT01564368), which ran as a substudy inside the I-SPY 2 neoadjuvant breast cancer trial. It tested whether the change in tumor apparent diffusion coefficient (ADC) during neoadjuvant chemotherapy predicts pathologic complete response (pCR). Women were enrolled at ten US institutions between August 2012 and January 2015. The University of California, San Francisco processed the images, and The Cancer Imaging Archive released the collection in 2021 under CC BY 4.0. It serves work on treatment response prediction, breast DWI analysis and ADC repeatability.

Composition

The collection holds 385 of the 388 eligible patients, all women aged 23 to 77 (median 49), with 1,123 MRI studies. Every patient has a pre-treatment study (T0); later studies were made only for the 272 patients randomized into I-SPY 2 treatment arms: early treatment after 3 weeks of paclitaxel (T1), mid-treatment between paclitaxel and anthracycline (T2) and post-treatment before surgery (T3). Download subsets cover the 242-patient primary analysis cohort, 71 patients with analyzable same-session test-retest DWI, and the 117 training and 74 test patients of the BMMR2 challenge. A spreadsheet adds age, race, tumor type, HR/HER2 subtype, grade, tumor size, pCR and per-study DWI quality ratings.

Acquisition

Scans were done prone on 1.5 T or 3 T scanners with a dedicated bilateral breast coil, the same scanner configuration for every study of a patient. The protocol required axial T2-weighted imaging, fat-suppressed single-shot echo-planar DWI with b = 0, 100, 600 and 800 s/mm² and T1-weighted DCE. GE, Siemens and Philips scanners appear in the metadata.

Annotations

For DWI studies that passed quality control, the collection includes trace images, ADC maps and manually drawn multi-slice whole-tumor segmentations from the trial analysis, as DICOM SEG and as MR objects. DCE studies come with ipsilateral cropped images, percent enhancement and signal enhancement ratio maps, and the functional tumor volume analysis mask.

Known limitations

• Only primary-analysis and test-retest studies are guaranteed to have segmentations and DWI derived objects.
• Some derived objects are not strictly DICOM compliant, and DCE functional tumor volumes stored here differ slightly from those used in I-SPY 2.
• Clinical data are limited because I-SPY 2 was ongoing at release.
• The 116 non-randomized patients have only a baseline study.