Overview

Schema Version

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

Name

VA National Teleradiology Program Noncontrast Head CT for AI ICH Evaluation (2021–2024)

Link

https://doi.org/10.1148/ryai.240067

Indexing

Keywords: Intracranial hemorrhage, noncontrast head CT, teleradiology, radiologist read time, report turnaround time, Avicenna.ai CINA v1.0, false positives, false negatives, system efficiency

Content: NR, CT

RadLex: RID45976, RID35976, RID45946, RID28768, RID4710, RID14

SNOMED: 1386000

Author(s)

Andrew James Del Gaizo, MD, MBA

Thomas F. Osborne, MD

Troy Shahoumian, MPH, PhD

Robert Sherrier, MD

Organization(s)

VA National Teleradiology Program

VA Palo Alto Health Care System

Stanford University School of Medicine, Department of Radiology

VA Health Solutions, Patient Care Services

Contact

Andrew James Del Gaizo (corresponding author)

Funding

No industry support; authors declared no funding.

Ethical review

Determination of quality assessment and nonresearch by the U.S. Department of Veterans Affairs; Stanford IRB determination of nonresearch, protocol 73642.

Comments

Retrospective evaluation of an AI clinical decision support tool (CINA v1.0, Avicenna.ai) for detecting acute intracranial hemorrhage on immediate noncontrast head CTs interpreted within a large VA national teleradiology program. Includes analysis of AI performance and impact on radiologist read times.

Date

Published: 2024-07-17

References

[1] Del Gaizo AJ, Osborne TF, Shahoumian T, Sherrier R. "Deep Learning to Detect Intracranial Hemorrhage in a National Teleradiology Program and the Impact on Interpretation Time". Radiology: Artificial Intelligence. 2024-09-01. doi:10.1148/ryai.240067. PMID: 39017032. PMCID: PMC11427938.

Dataset

Motivation

Assess diagnostic performance of an ICH AI tool in a low-prevalence, high-volume teleradiology setting and quantify impact on interpretation times/system efficiency.

Sampling

Consecutive immediate noncontrast head CT examinations interpreted by the VA National Teleradiology Program during specified periods.

Partitioning scheme

Pre-AI baseline period (Aug 2021–May 2022) vs Post-AI period (Jan 2023–Feb 2024). Within post-AI, AI outcomes categorized as TP, FP, TN, FN; AI error outputs excluded from performance analysis.

Missing information

Patient demographics, age/sex distributions, and image acquisition parameters per scanner not detailed in the article text.

Relationships between instances

Each CT exam is linked to an AI output (screen capture) and the corresponding radiologist report used as reference standard.

Noise

AI error outputs occurred on 5.48% (3383/61704) of post-AI studies; false positives associated with artifacts, infarct, calcifications, postoperative changes, and unknown/indeterminate causes.

External data

None reported.

Confidentiality

Clinical VA teleradiology data; retrospective quality assessment/nonresearch. Data sharing upon request; no individual de-identified participant data will be shared.

Sensitive data

Patient imaging and reports from a federal healthcare system; potentially sensitive PHI though not shared publicly.