Overview

Schema Version

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

Name

Acute/Subacute Intracranial Hemorrhage on Noncontrast Head CT

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9152881/

Indexing

Keywords: noncontrast head CT, intracranial hemorrhage, acute hemorrhage, subacute hemorrhage, hemorrhage subtyping, segmentation, confidence scores, Dempster-Shafer, calibrated classifier, worklist prioritization

Content: CT, NR, ER

RadLex: RID28768, RID4710

SNOMED: 1386000, 35486000, 82999001, 21454007, 23276006

Author(s)

Eli Gibson

Bogdan Georgescu

Pascal Ceccaldi

Pierre-Hugo Trigan

Youngjin Yoo

Jyotipriya Das

Thomas J. Re

Vishwanath RS

Abishek Balachandran

Eva Eibenberger

Andrei Chekkoury

Barbara Brehm

Uttam K. Bodanapally

Savvas Nicolaou

Pina C. Sanelli

Thomas J. Schroeppel

Thomas Flohr

Dorin Comaniciu

Yvonne W. Lui

Organization(s)

Siemens Healthineers, Department of Digital Technology and Innovation

Siemens Healthineers, Bangalore, India

Siemens Healthineers, Department of Computed Tomography

University of Maryland Medical Center, Department of Radiology

Vancouver General Hospital, Department of Radiology

Northwell Health, Department of Radiology

UCHealth Memorial Hospital, Department of Surgery

NYU Langone Health, Department of Radiology, New York University School of Medicine

License

URL: https://pubs.rsna.org/doi/10.1148/ryai.210115

Contact

Corresponding author: Eli Gibson (moc.sreenihtlaeh-snemeis@nosbig.ile)

Funding

Financially supported by Siemens Healthineers.

Ethical review

Retrospective, anonymized NCCT head volumes collected from 10 centers with ethics committee approval and waiver of informed consent; HIPAA compliant.

Date

Published: 2022-04-20

References

[1] Flanders AE, Prevedello LM, Shih G, et al.. "Construction of a machine learning dataset through collaboration: the RSNA 2019 Brain CT Hemorrhage Challenge". Radiology: Artificial Intelligence. 2020-01-01. PMID: 33937827. PMCID: PMC8082297. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8082297/

Dataset

Motivation

Evaluate AI that detects, subtypes, and locates ICH on NCCT head scans and produces statistical confidence scores to identify high-confidence subsets and improve simulated worklist prioritization.

Sampling

Retrospective collection from 10 centers (United States, Brazil, Canada, India). Inclusion: adult patients (>18 years); exclusion: age ≤18, incompatible image geometry, or absence of axial reconstruction.

Partitioning scheme

Internal centers randomly split 9:1 at patient level once before development; external (RSNA) data randomly split once to obtain a small calibration set; fixed internal vs external split.

Missing information

Subtype labels unavailable for 1753 of 9306 ICH-positive development studies and 115 of 1032 ICH-positive internal evaluation studies (excluded from subtype performance). Manual segmentations available for a subset (3278/10,038 ICH-positive internal-center studies).

Relationships between instances

Multiple hemorrhage subtypes may co-occur within a study; independent subtype labels allow multi-label classification per study.

Noise

Data from multiple scanner manufacturers (Siemens Healthineers, GE Healthcare, Toshiba); varying acquisition parameters across 10 centers.

External data

Includes studies from the RSNA ICH Detection Challenge training dataset (used as external centers).

Confidentiality

Anonymized datasets; HIPAA compliant.

Re-identification

Data were anonymized; centers ensured de-identification and compliance with local regulations.

Sensitive data

Clinical radiology reports used for label derivation where available; PHI removed during anonymization.