Related Experiment Video
Updated: Jul 12, 2025
08:49
Whole-Brain Single-Cell Imaging and Analysis of Intact Neonatal Mouse Brains Using MRI, Tissue Clearing, and Light-Sheet Microscopy
Published on: August 1, 2022
3.6K
Mu-Net a Light Architecture for Small Dataset Segmentation of Brain Organoid Bright-Field Images.
Clara Brémond Martin1,2, Camille Simon Chane1, Cédric Clouchoux2
1ETIS Laboratory UMR 8051 (CY Cergy Paris Université, ENSEA, CNRS), 6 Avenue du Ponceau, 95000 Cergy, France.
Biomedicines
|October 28, 2023
Summary
Researchers developed MU-Net, a deep learning model for segmenting brain organoids (BOs). This light U-Net architecture effectively analyzes small datasets, offering robust segmentation for brain organoid growth characterization.
Area of Science:
- Neuroscience
- Biotechnology
- Computer Science
Background:
- Manual segmentation of brain organoids (BOs) is time-consuming.
- Deep learning (DL) for segmentation requires large datasets, which are scarce for novel BO cultures.
- Light U-Net architectures offer a solution for small datasets.
Purpose of the Study:
- To develop and evaluate a novel, lightweight U-Net architecture (MU-Net) for segmenting brain organoids.
- To compare MU-Net's performance against existing U-Net models (U-Net, UNet-Mini) using limited brain organoid image data.
- To assess the robustness of MU-Net across different data augmentation strategies.
Main Methods:
- Proposed a novel, reduced U-Net architecture named MU-Net.
- Compared MU-Net with U-Net and UNet-Mini on bright-field brain organoid images.
- Utilized leave-one-out cross-validation with original, synthesized (via adversarial autoencoder), and transformed image datasets.
- Employed various data augmentation strategies to train the models.
Main Results:
- U-Net with optimized augmentation yielded the best segmentation results.
- The novel MU-Net demonstrated high robustness, achieving comparable segmentation accuracy across diverse training datasets.
- MU-Net proved effective even with limited or varied data, outperforming other light U-Net models in consistency.
Conclusions:
- Lightweight U-Net methods, including the proposed MU-Net, can effectively segment brain organoids from small datasets.
- MU-Net offers a robust and accurate alternative for brain organoid image segmentation, particularly valuable in research with limited data.
- This study validates the feasibility of automated segmentation for characterizing brain organoid development using deep learning.
