Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

Scottish Breast Screening Service (SBSS) mammography dataset (2016–2019)

Link

https://dx.doi.org/10.1148/ryai.220146

Indexing

Keywords: Breast, Screening, Mammography, Computer Applications–Detection/Diagnosis, Neoplasms-Primary, Technology Assessment

Content: BR, RS

RadLex: RID29896, RID29897, RID10357

SNOMED: 254837009

Author(s)

Clarisse F. de Vries

Samantha J. Colosimo

Roger T. Staff

Jaroslaw A. Dymiter

Joseph Yearsley

Deirdre Dinneen

Moragh Boyle

David J. Harrison

Lesley A. Anderson

Gerald Lip

Organization(s)

University of Aberdeen

NHS Grampian

Kheiron Medical Technologies

Canon Medical Research Europe (SHAIP platform)

University of St Andrews

License

Text: Data access subject to required approvals and data agreements via Grampian Data Safe Haven (DaSH).

URL: https://www.abdn.ac.uk/iahs/facilities/grampian-data-safe-haven.php

Contact

Corresponding author: Clarisse F. de Vries (ku.ca.ndba@seirved.essiralc)

Funding

Supported by the Industrial Centre for Artificial Intelligence Research in Digital Diagnostics (iCAIRD), funded by Innovate UK on behalf of UK Research and Innovation (UKRI) (project no. 104690).

Ethical review

Approved by London-Bloomsbury Research Ethics Committee (20/LO/0563); Public Benefit and Privacy Panel approval (1920–0258). Secondary use of de-identified data; individual consent not required.

Comments

Retrospective unenriched screening dataset consecutively acquired from a U.K. regional screening program; used to evaluate generalizability and threshold calibration of a commercial AI (Mia).

Date

Published: 2023-03-22

Created: 2016-04-01

References

[1] de Vries CF; Colosimo SJ; Staff RT; et al.. "Impact of Different Mammography Systems on Artificial Intelligence Performance in Breast Cancer Screening". Radiology: Artificial Intelligence. 2023-05-01. doi:10.1148/ryai.220146. PMID: 37293340. PMCID: PMC10245180.

Dataset

Motivation

Evaluate generalizability of a commercial AI across different mammography system software versions and calibrate site-specific decision thresholds.

Sampling

Consecutively acquired routine screening attendees in a U.K. regional program over one 3-year screening cycle; exclusions per vendor recommendations and service criteria.

Partitioning scheme

Original dataset (after exclusions), validation dataset for threshold generation, test dataset (hold-out) for evaluation; additional post-upgrade positives added for threshold robustness.

Missing information

Exact image resolution and pixel spacing not reported; racial/ethnic breakdown largely unavailable; detailed DICOM tag lists not provided.

Relationships between instances

Standard screening exam comprises four image views per attendee (CC and MLO for each breast).

Noise

Variability introduced by mammography unit software versions (1.7 vs 1.8) affected recall rates.

External data

Clinical outcomes linked in DaSH; AI outputs (malignancy prediction values) generated within SHAIP without vendor access to outcomes.

Confidentiality

De-identified within Safe Haven environments (DaSH, SHAIP) using “Hiding in Plain Sight” de-identification.

Re-identification

Mitigated through Safe Haven use and text de-identification ('Hiding in Plain Sight').

Sensitive data

Contains medical imaging and linked clinical outcomes; access requires approvals and data agreements via DaSH.