Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 29 of the 29 references it cites were found in a registry or an index; 5 of the 21 sources its claims cite were checked for support.

3D MRI Brain Tumor Segmentation Using Autoencoder Regularization

Open access
Myronenko, Andriy · 2019 · Lecture Notes in Computer Science

Overview

CoChat Summary

TL;DR

The paper describes a 3D MRI brain tumor segmentation network with an encoder-decoder architecture and a variational autoencoder branch used only during training to regularize the shared encoder. It won 1st place in the BraTS 2018 challenge, reporting testing Dice of 0.7664 for enhancing tumor, 0.8839 for whole tumor, and 0.8154 for tumor core with an ensemble of 10 models.

Key findings

  • The proposed method won 1st place on the BraTS 2018 testing dataset (191 cases); an ensemble of 10 models achieved mean Dice of 0.7664 (ET), 0.8839 (WT), and 0.8154 (TC), with Hausdorff distances of 3.7731 mm (ET), 5.9044 mm (WT), and 4.8091 mm (TC).
  • On the BraTS 2018 validation dataset (66 cases), a single model achieved Dice of 0.8145 (ET), 0.9042 (WT), and 0.8596 (TC); test-time augmentation only marginally improved performance, while an ensemble of 10 models improved results by about 1% to Dice of 0.8233 (ET), 0.9100 (WT), and 0.8668 (TC).
  • The VAE branch, used only during training, regularized the shared encoder and improved performance, helping to consistently achieve good training accuracy for any random initialization.
  • Training for 300 epochs took 2 days on a single NVIDIA Tesla V100 32GB GPU, or 6 hours on an NVIDIA DGX-1 with 8 V100 GPUs; inference time was 0.4 sec for a single model on a single V100 GPU.

Methods

  • The network uses an encoder-decoder CNN with ResNet-like blocks, Group Normalization, 3x3x3 convolutions, and an initial 32 filters; the input is a four-channel 3D MRI crop, and the segmentation decoder outputs three nested tumor subregions (WT, TC, ET) after a sigmoid.
  • A VAE branch reconstructs the input image from the encoder endpoint; the loss combines soft Dice loss, L2 reconstruction loss, and KL divergence, with a weight of 0.1 balancing Dice and VAE terms. Training used Adam with initial learning rate 1e-4, batch size 1, 300 epochs, and random crops of 160x192x128.
  • Data were from BraTS 2018: 285 training cases (210 HGG, 75 LGG), 66 validation cases, and 191 testing cases, each with four aligned MRI modalities (T1, T1c, T2, FLAIR). Preprocessing included zero-mean/unit-std normalization, random intensity shift and scale, and random axis mirror flips; evaluation used test-time augmentation and an ensemble of 10 models.

Limitations

  • The paper states that the training dataset size was limited, which motivated adding the VAE branch for regularization; no additional in-house training data were used.
  • The paper states that GPU memory limits constrained the batch size to 1 and the crop size; a larger batch size of 8 to use BatchNorm required a smaller image crop and resulted in worse performance.
  • Evident from the text: evaluation is limited to the BraTS 2018 validation and testing datasets, with no external or in-house validation reported; several attempted alternatives (CRF post-processing, more sophisticated augmentation, increased network depth) did not improve results.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/978-3-030-11726-9_28

Cites

29 of 29 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

1 Valid · 1 Not valid · 5 Not checked · 18 Own results and interpretations

11 more claims

Create a free account to see all 25 claims and to verify the 16 references still open.

See all claims — free account
  • Probably not a substantial claim. 5
  • It may state two separate claims as one. 8
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 23 references

Cited by

See all 1,283

Metadata

Authors
Myronenko, A.
Affiliations
Nvidia (United States) (US)
Corresponding author
Myronenko, Andriy
Venue
Lecture Notes in Computer Science
Published
2019
Pages
311-320
ISSN
  • 0302-9743 (ISSN-L)
  • 1611-3349
OpenAlex
W2898522275
Cited by
1,283
References
23
Citations by year
2026: 163 · 2025: 208 · 2024: 205 · 2023: 165 · 2022: 161 · 2021: 198 · 2020: 136 · 2019: 46
FWCI
79.33
Type
conference paper
Language
en
Open access
Licence
Locations
  • Lecture notes in computer science (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
  • TIB Data Manager
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Brain Tumor Detection and Classification
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
  • Health Sciences › Medicine › Oncology › Glioma Diagnosis and Treatment
Keywords
brain tumor segmentation, 3D MRI, semantic segmentation, variational autoencoder, encoder-decoder architecture, BraTS 2018, limited training data

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: 3D MRI Brain Tumor Segmentation Using Autoencoder Regularization. https://app.cochat.ai/papers/10.1007/978-3-030-11726-9_28
The paper
Myronenko, A. (2019). 3D MRI Brain Tumor Segmentation Using Autoencoder Regularization. Lecture Notes in Computer Science. https://doi.org/10.1007/978-3-030-11726-9_28

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.