What question did this study set out to answer?

To introduce and validate FatSegNet, an automated deep learning pipeline designed to identify, segment, and quantify visceral and subcutaneous abdominal adipose tissue on Dixon MRI scans.

October 21, 2019Magnetic Resonance in MedicineOpen Access

FatSegNet: A fully automated deep learning pipeline for adipose tissue segmentation on abdominal dixon MRI

View Full Paper

Ask AI

Bookmark

Why the study?

To introduce and validate a fast, fully automated deep learning pipeline (FatSegNet) to accurately identify, segment, and quantify visceral and subcutaneous adipose tissue within a consistent abdominal region on Dixon MRI scans.

Key Points

To introduce and validate FatSegNet, an automated deep learning pipeline designed to identify, segment, and quantify visceral and subcutaneous abdominal adipose tissue on Dixon MRI scans.
FatSegNet was structured into three stages: abdominal localization using two 2D-Competitive Dense Fully Convolutional Networks (CDFNet), multi-view fat segmentation with independent CDFNets, and multi-view aggregation.
The pipeline was evaluated via sixfold cross-validation, test-retest reliability, manual re-editing generalizability, and replication of age and sex effects within the Rhineland Study cohort.
FatSegNet achieved a Dice similarity coefficient of 0.850 for visceral adipose tissue (VAT) compared to 0.788 for manual raters, and 0.975 for subcutaneous adipose tissue (SAT) compared to 0.982.
Reliability assessments demonstrated high agreement, with test-retest intraclass correlation coefficients (ICC) of 0.998 for VAT and 0.996 for SAT, and manual re-editing ICCs of 0.999 for both tissues.
The fully automated pipeline processed a complete 3D Dixon MRI scan in approximately one minute.

Structured PICO

Population

Abdominal Dixon MRI scans from the Rhineland Study (a large prospective population cohort)

Intervention

FatSegNet (a fully automated deep learning pipeline using 2D-Competitive Dense Fully Convolutional Networks)

Comparator

Manual raters and traditional deep learning networks

Outcome

Segmentation accuracy (Dice score) for visceral and subcutaneous adipose tissue (VAT and SAT)surrogate

FatSegNet provides a fast, fully automated, and highly accurate deep learning pipeline for segmenting abdominal visceral and subcutaneous adipose tissue on Dixon MRI.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

May enable scalable VAT segmentation in research cohorts; leaves open prospective validation before clinical adoption.

Cite This Study

A 2019 study studied this question.

synapsesocial.com/papers/6a886aa3ce59b4d55bcf187f https://doi.org/10.1002/mrm.28022

View Full Paper

Ask AI

Bookmark

FatSegNet: A fully automated deep learning pipeline for adipose tissue segmentation on abdominal dixon MRI

Key Points

Structured PICO

Discussion

Implication

Cite This Study

FatSegNet: A fully automated deep learning pipeline for adipose tissue segmentation on abdominal dixon MRI

Key Points

Structured PICO

Discussion

Implication

Cite This Study

Keep reading

Keep reading