Overview

Schema Version

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

Name

HD-SEQ-ID

Link

https://www.github.com/neuroAI-HD/HD-SEQ-ID

Indexing

Keywords: MR-Imaging, Neural Networks, CNS, Brain/Brain Stem, Computer Applications-General (Informatics), Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms

Content: MR, NR

RadLex: RID35806, RID28694, RID12698, RID10312, RID39467, RID10799, RID49531, RID10796

SNOMED: 1163375002

Author(s)

Mustafa Ahmed Mahmutoglu, MD

Chandrakanth Jayachandran Preetha, MSc

Hagen Meredig, MD

Joerg-Christian Tonn, MD

Michael Weller, MD

Wolfgang Wick, MD

Martin Bendszus, MD

Gianluca Brugnara, MD

Philipp Vollmuth, MD, MBA

Organization(s)

Heidelberg University Hospital, Department of Neuroradiology

Heidelberg University Hospital, Division for Computational Neuroimaging

Heidelberg University Hospital, Department of Neurology

University Hospital Munich LMU, Department of Neurosurgery

University Hospital Zurich, Department of Neurology, Clinical Neuroscience Center

University of Zurich

European Organization for Research and Treatment of Cancer (EORTC)

Version

1.0

Contact

ed.grebledieh-inu.dem@htumllov.ppilihp

Funding

Parts of the MRI data from CORE (Merck KGaA), CENTRIC (EORTC supported by Merck KGaA), and EORTC 26101 (supported by Hoffmann-La Roche). Supported by Deutsche Forschungsgemeinschaft (DFG) 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 Else Kröner Clinician Scientist Endowed Professorship (Else Kröner Fresenius-Stiftung).

Ethical review

Retrospective study. Institutional cohort approved by local ethics committee (Heidelberg University Hospital; reference S-784/2018). Evaluation of CENTRIC, CORE, and EORTC-26101 cohorts granted via EORTC external research projects (references 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 Jan;6(1):e230095. Published online 2023 Nov 15.. . doi:10.1148/ryai.230095. PMID: 38166331. PMCID: PMC10831512.

Model

Architecture

Convolutional Neural Network based on ResNet-18; ensemble across five folds. A ResNet-50 variant was also evaluated.

Availability

Open-source with pretrained models and documentation at https://www.github.com/neuroAI-HD/HD-SEQ-ID

Clinical benefit

Automated labeling of brain MRI sequence types to enhance speed, accuracy, and efficiency of clinical and research neuroradiologic workflows; potential integration into AI-assisted hanging protocols.

Clinical workflow phase

Workflow optimization; image organization for interpretation (hanging protocols); data curation and preprocessing for downstream analysis.

Degree of automation

Fully automated sequence-type classification from images.

Indications for use

Identification of nine MRI sequence types in brain MRI datasets from heterogeneous, multicenter sources (intended for research data curation and clinical workflow support in neuroradiology).

Input

Two-dimensional midsection images extracted from volumetric brain MRI sequences (T1w, postcontrast T1w, T2w, FLAIR, SWI, ADC, DWI low b, DWI high b, T2*/DSC-related).

Instructions

Provide 2D midsection images derived from each MRI series as input as described; see project documentation at the GitHub repository for preprocessing and usage details.

Limitations

Underrepresentation of SWI (only 878 images; 16 in test set) may impact performance for SWI. Training cohorts consisted of glioblastoma patients; although analyses suggest generalizability, additional external validation is encouraged. Possible confusion between T1w and postcontrast T1w when midsection lacks enhancement landmarks; performance may vary with scanner type and contrast agent concentration. Approach uses a single 2D midsection from 3D volumes, which may omit informative anatomy. More advanced architectures might further improve performance.

Output

CDEs: RDE2006.1, RDE1545, RDE1453, RDE627, RDE1832, RDE2006.0, RDE1452

Description: Multiclass classification label identifying the MRI sequence type.

Recommendation

Suitable for automated sequence labeling and integration with hanging protocols to reduce manual adjustments in neuroradiologic workflow.

Regulatory information

Authorization status: Not a regulated medical device (research software); no regulatory clearance reported.

Reproducibility

Code and pretrained models publicly available; training/validation/test stratified fivefold split described; detailed preprocessing and metrics in supplemental material.

Use

Intended: Image processing

Out-of-scope: Other

Excluded: Diagnosis

User

Intended: Radiologist, Subspecialist diagnostic radiologist, Researcher

Out-of-scope: Layperson