Overview

Schema Version

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

Name

Virginia Mason Franciscan Health screening mammography (2D FFDM) quality-improvement dataset

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10698591/

Indexing

Keywords: mammography, screening, image quality, breast positioning, compression pressure, technical repeat, technical recall, EQUIP, MQSA, Volpara Analytics

Content: BR, SQ, IN

RadLex: RID10569, RID49614, RID45974, RID28816, RID28657, RID14, RID10357

Author(s)

Peter R. Eby

Linda M. Martis

Jeremy T. Paluch

Janice J. Pak

Ariane H. L. Chan

Organization(s)

Virginia Mason Franciscan Health

Volpara Health Technologies

Funding

Authors declared no funding for this work.

Ethical review

Institutional review determined this retrospective analysis satisfied Quality Improvement Project requirements and did not constitute human subjects research; IRB waiver granted. AI software data considered de-identified via expert determination under HIPAA.

Comments

Retrospective institutional dataset used to assess impact of AI software (Volpara Analytics) on image quality metrics and technical repeat/recall rates across nine mammography facilities.

Date

Published: 2023-10-25

References

[1] Eby PR, Martis LM, Paluch JT, Pak JJ, Chan AHL. "Impact of Artificial Intelligence–driven Quality Improvement Software on Mammography Technical Repeat and Recall Rates". Radiology: Artificial Intelligence. 2023-11-01. doi:10.1148/ryai.230038. PMID: 38074792. PMCID: PMC10698591.

Dataset

Motivation

Assess whether AI software implementation improved objectively measured image quality and reduced technical repeats/recalls.

Sampling

Full-field 2D screening examinations from April 2019 to March 2022 across nine facilities; diagnostic, synthetic, tomosynthesis, and implant examinations excluded.

Partitioning scheme

Authors defined datasets 1A/2A (full range), 1B/2B (baseline vs current periods, excluding middle year), and 1C/2C (common technologists) for analysis.

Missing information

Some images lacked IQ and/or breast composition metrics due to algorithm sanity check failures or incorrect/missing DICOM header information; treated as missing.

Relationships between instances

Images are grouped within examinations; examinations acquired by specific technologists across nine facilities; IQ metrics computed per image and aggregated per technologist/time period.

External data

No external datasets; data derived from institutional clinical systems (Centricity) and AI software (Volpara Analytics).

Confidentiality

AI software data considered de-identified via expert determination under HIPAA.

Re-identification

Data de-identified via expert determination; no patient-identifying information reported.

Sensitive data

Breast imaging data; de-identified per HIPAA expert determination.