Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Jun 30, 2026

Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules

Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules

Published on: October 13, 2023

MAE-UNETR++: Masked Autoencoder Pretraining for 3-D Lung Nodule Segmentation.

Vinayak Savant, Yue Wang, Jianhua Xuan

Biorxiv : the Preprint Server for Biology

|June 29, 2026

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Molecular mechanism leading to human coronary atherosclerosis assessed by proteomic analysis and RNA sequences.

European heart journal·2026

Same author

Impact of common variants on brain gene expression from RNA to protein to schizophrenia risk.

Nature communications·2025

Same author

Precision and Accuracy in Quantitative Measurement of Gene Expression from Single-cell/nucleus RNA Sequencing Data.

Genomics, proteomics & bioinformatics·2025

Same author

Prediction of gene regulatory connections with joint single-cell foundation models and graph-based learning.

Bioinformatics (Oxford, England)·2025

Same author

iDDN: determining trans-omics network structure and rewiring with integrative differential dependency networks.

Bioinformatics advances·2025

Same author

Prediction of Gene Regulatory Connections with Joint Single-Cell Foundation Models and Graph-Based Learning.

bioRxiv : the preprint server for biology·2025

Same journal

A human-specific genetic modifier reconfigures large-scale cortical network dynamics underlying behavioral performance.

bioRxiv : the preprint server for biology·2026

Same journal

<i>Staphylococcus aureus</i> uses a eukaryotic-like uridyltransferase to make UDP-GlcNAc for cell wall synthesis.

bioRxiv : the preprint server for biology·2026

Same journal

Dynamic redistribution of eIF4F controls cap-dependent translation initiation.

bioRxiv : the preprint server for biology·2026

Same journal

When does additional information improve accuracy of RNA secondary structure prediction?

bioRxiv : the preprint server for biology·2026

Same journal

Normative brain-state trajectories reveal deviation from healthy aging in Alzheimer's disease.

bioRxiv : the preprint server for biology·2026

Same journal

Noradrenergic infraslow rhythm during sleep is the critical link between heart-rate dynamics and memory consolidation.

bioRxiv : the preprint server for biology·2026

See all related articles

Masked autoencoder (MAE) pretraining improves 3-D lung nodule segmentation accuracy, outperforming random initialization and standard transfer learning. This self-supervised learning approach enhances model performance, especially with limited labeled data.

Area of Science:

Medical Imaging
Artificial Intelligence
Computer Vision

Background:

Voxel-level annotation for 3-D medical imaging is costly and hard to scale.
Training high-capacity 3-D segmentation models is challenging due to data limitations.
Transfer learning (TL) can underperform when source and target domains differ, particularly for pulmonary nodules.

Purpose of the Study:

To propose and evaluate a masked autoencoder (MAE) pretraining approach for domain-specific self-supervised learning (SSL).
To address data efficiency challenges in 3-D lung nodule segmentation caused by domain differences.
To investigate the effectiveness of MAE pretraining compared to random initialization and existing TL methods.

Main Methods:

Implemented MAE pretraining on target-domain CT volumes for 3-D lung nodule segmentation.

More Related Videos

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules

A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules

Published on: May 19, 2023

Related Experiment Videos

Last Updated: Jun 30, 2026

Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules

Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules

Published on: October 13, 2023

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules

A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules

Published on: May 19, 2023

Evaluated MAE pretraining against random initialization using baseline models.

Compared MAE pretraining with Decathlon TL for the UNETR++ model and assessed its impact on a CNN baseline (V-Net).

Main Results:

MAE pretraining achieved a Dice Similarity Coefficient (DSC) of 0.307, surpassing random initialization (0.136) and Decathlon TL (0.257).
MAE pretraining significantly improved V-Net stability in low-data regimes, increasing DSC from 0.010 to 0.071.
Results demonstrate MAE's effectiveness in enhancing segmentation performance with limited labeled data.

Conclusions:

MAE-based pretraining offers a practical and robust initialization strategy for volumetric segmentation tasks.
This self-supervised learning method effectively overcomes domain differences in medical imaging datasets.
MAE pretraining is particularly beneficial for 3-D lung nodule segmentation when labeled data are scarce.