Overview

Schema Version

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

Name

Multicenter glioblastoma brain MRI cohort for HD-SEQ-ID (CENTRIC, CORE, EORTC-26101, Heidelberg)

Link

https://doi.org/10.1148/ryai.230095

Indexing

Keywords: brain MRI, sequence identification, multicenter, glioblastoma, ResNet-18, HD-SEQ-ID, DWI, FLAIR, ADC, SWI, DSC, GRE T2*

Content: MR, NR, OI

RadLex: RID35976, RID35806, RID12698, RID10312, RID39536, RID4044, RID49531, RID10796

SNOMED: 1163375002

Author(s)

Mustafa Ahmed Mahmutoglu

Chandrakanth Jayachandran Preetha

Hagen Meredig

Joerg-Christian Tonn

Michael Weller

Wolfgang Wick

Martin Bendszus

Gianluca Brugnara

Philipp Vollmuth

Organization(s)

Heidelberg University Hospital

European Organization for Research and Treatment of Cancer (EORTC)

University Hospital Munich LMU

University Hospital Zurich and University of Zurich

Funding

Parts of the MRI data acquired through Merck KGaA–supported CENTRIC and CORE studies and the EORTC 26101 study supported by Hoffmann-La Roche. Additional support: Deutsche Forschungsgemeinschaft (project identifiers 404521405 [SFB 1389—UNITE Glioblastoma, WP C02] and 428223917 [Priority Programme 2177 Radiomics: MA 6340/18–2, VO 2801/1–2]). P.V. supported by an Else Kröner Clinician Scientist Endowed Professorship (Else Kröner Fresenius-Stiftung).

Ethical review

Institutional cohort approved by local ethics committee (Heidelberg, reference S-784/2018). Evaluation of CENTRIC, CORE, and EORTC-26101 cohorts granted through EORTC external research projects (ERP-263 and ERP-362).

Date

Published: 2023-11-15

References

[1] Mahmutoglu MAM, Preetha CJ, Meredig H, Tonn JC, Weller M, Wick W, Bendszus M, Brugnara G, Vollmuth P. "Deep Learning–based Identification of Brain MRI Sequences Using a Model Trained on Large Multicentric Study Cohorts". Radiology: Artificial Intelligence. 2024-01-01. doi:10.1148/ryai.230095. PMID: 38166331. PMCID: PMC10831512.

Dataset

Motivation

To enable fully automated, reliable labeling of heterogeneous multicenter MRI sequences for clinical and research workflows.

Sampling

Retrospective, multicenter sampling from 249 hospitals and 29 scanner models (1–3 T) across glioblastoma cohorts.

Partitioning scheme

Stratified fivefold split balanced across institutions, patients, and sequence types. One fold used as test (~20%); remaining ~80% split again into training and validation using the same stratified fivefold approach (approx. 64% train, 16% validation).

Relationships between instances

Multiple examinations per patient; multiple sequences per examination; two-dimensional midsection image extracted per sequence for model input.

Noise

Bad-quality MRI data were visually excluded (~15%, 9618 MR images).

External data

Four cohorts: institutional Heidelberg cohort; clinical trials CENTRIC (NCT00689221), CORE (NCT00813943), and EORTC-26101 (NCT01290939).

Sensitive data

Medical imaging of patients with histologically confirmed glioblastoma.