Overview

Schema Version

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

Name

University of Wisconsin–Madison FDG PET/CT lymphoma reports and images (2008–2018)

Link

https://dx.doi.org/10.1148/ryai.220281

Indexing

Keywords: Deauville score, PET/CT, FDG, Lymphoma, Radiology reports, Masked language modeling, Domain adaptation, Wisconsin

Content: NM, OI, IN

RadLex: RID35976, RID11701, RID3842, RID10341, RID12782

Author(s)

Zachary Huemann

Changhee Lee

Junjie Hu

Steve Y. Cho

Tyler J. Bradshaw

Organization(s)

University of Wisconsin–Madison

University of Wisconsin Carbone Cancer Center

Contact

ude.csiw@nnameuhz

Funding

Supported by GE HealthCare; NVIDIA provided an RTXA6000 GPU to the author’s institution.

Ethical review

Institutional review board–approved, retrospective, HIPAA-compliant protocol with waiver of informed consent.

Comments

Retrospective single-institution study of PET/CT examinations and associated radiology reports focused on Deauville score prediction.

Date

Published: 2023-09-27

References

[1] Huemann Z, Lee C, Hu J, Cho SY, Bradshaw TJ. "Domain-adapted Large Language Models for Classifying Nuclear Medicine Reports". Radiology: Artificial Intelligence. 2023-11-01. doi:10.1148/ryai.220281. PMID: 38074793. PMCID: PMC10698610.

Dataset

Motivation

Evaluate impact of domain adaptation on language models for predicting Deauville scores from PET/CT reports.

Sampling

Queried PACS for clinical 18F-FDG PET/CT examinations containing the term “lymphoma” in the indication or impression.

Partitioning scheme

Seven iterations of random-sampling cross-validation with 80% training, 10% validation, 10% test on the 1,664 labeled examinations.

Missing information

Deauville scores absent for a subset of examinations acquired prior to adoption of Deauville criteria or with non-lymphoma indications.

Relationships between instances

Examination-level dataset; potential multiple examinations per patient (patient-level counts not reported).

Noise

Labels are physician-assigned Deauville scores extracted from reports; if multiple DSs present, highest value used; potential inter-physician variability.

External data

None reported beyond the single-institution PACS-derived dataset.

Confidentiality

De-identified clinical imaging data and reports; HIPAA-compliant.

Re-identification

Images anonymized using Clinical Trial Processor (RSNA CTP) to remove protected health information.

Sensitive data

Clinical images and reports; PHI removed prior to analysis.