Overview

Schema Version

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

Name

UK Biobank neck-to-knee body MRI scans (MIMIR study subset)

Link

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7250878/

Indexing

Keywords: UK Biobank, neck-to-knee MRI, body composition, organ volumes, liver fat, 1.5-T MRI, mean-variance regression, deep learning

Content: MR, BQ

RadLex: RID10840

Author(s)

Taro Langner

Andrés Martínez Mora

Robin Strand

Håkan Ahlström

Joel Kullberg

Organization(s)

Uppsala University

Antaros Medical AB

UK Biobank

Contact

Corresponding author: Taro Langner (email as provided in article: dev@null)

Funding

Supported by the Swedish Heart-Lung Foundation and the Swedish Research Council (2016-01040, 2019-04756, 2020-0500, 2021-70492) and the UK Biobank Resource under application no. 14237.

Ethical review

Covered by a Research Tissue Bank approval from UK Biobank and separate approval from the responsible Swedish ethics committee.

Comments

Retrospective cross-validation study using 38,916 UK Biobank 1.5-T neck-to-knee body MRI scans to train and validate deep regression models (MIMIR) estimating 72 measurements; subsequent evaluation against a later UK Biobank release for 12 body composition targets in 15,000 participants.

Date

Published: 2022-04-06

References

[1] Langner T, Martínez Mora A, Strand R, Ahlström H, Kullberg J. "MIMIR: Deep Regression for Automated Analysis of UK Biobank MRI Scans". Radiology: Artificial Intelligence. 2022. doi:10.1148/ryai.210178. PMID: 35652115. PMCID: PMC9152682.

[2] Littlejohns TJ, Holliday J, Gibson LM, et al.. "The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions". Nature Communications. 2020. PMID: 32457287. PMCID: PMC7250878. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7250878/

Dataset

Motivation

Automate estimation of 72 measurements (body composition, organ volumes, age/sex/anthropometrics, and selected experimental properties) from UK Biobank MRI to accelerate large-scale research.

Sampling

Three imaging centers; adult participants aged 44–82 years (mean 64), 52% female, BMI 14–62 kg/m2 (mean 27), 95% self-reported White British ethnicity.

Partitioning scheme

Stratified 10-fold cross-validation by grouping all participants into 10 even subsets; additional testing against a subsequent UKB release for 12 body composition targets in 15,000 participants.

Missing information

Across all participants, 73% of target values were not available through UKB; estimating these missing measurements was a key motivation.

Relationships between instances

One neck-to-knee MRI scan per participant processed into 2D projected representations; four multi-output network modules trained to predict different target groups.

Noise

Scans with water-fat signal swaps, corrupted data, implants, severe abnormalities (eg, large tumors), or nonstandard positioning were excluded after visual inspection.

Confidentiality

Data shared by UK Biobank under Research Tissue Bank approval; access governed by UK Biobank approvals.