Related Experiment Video
Updated: Jul 22, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.5K
Multi-class semantic segmentation of breast tissues from MRI images using U-Net based on Haar wavelet pooling
Kwang Bin Yang1, Jinwon Lee2, Jeongsam Yang3
1Devision of Memory - Memory FAB Team 1, Samsung Electronics, 1 Samsungjeonja-ro, Hwaseong, Gyeonggi, 18448, Republic of Korea.
Scientific Reports
|July 20, 2023
Summary
This study introduces a novel U-Net based method for segmenting breast tissues from MRI scans. This technique accurately reconstructs natural breast shapes, overcoming challenges posed by varying tissue elasticity.
Area of Science:
- Biomedical Imaging
- Medical Image Analysis
- Computational Biology
Background:
- Breast MRI scans, typically acquired in a supine position, do not accurately represent natural breast shape in an upright posture.
- Previous methods using finite element analysis struggle with heterogeneous breast tissue elasticity, limiting accurate shape reconstruction.
- Accurate breast shape reconstruction is crucial for various applications, including surgical planning and prosthetic design.
Purpose of the Study:
- To develop and validate a multi-class semantic segmentation method for precise breast tissue classification.
- To reconstruct the natural breast shape in a standing position using segmented MRI data.
- To improve the accuracy of breast shape modeling by addressing the challenge of varying tissue elastic moduli.
Main Methods:
- A dataset was curated and annotated with skin, fat, fibro-glandular tissue, and background labels from supine MRI scans.
- A U-Net architecture incorporating Haar wavelet pooling was employed for multi-class semantic segmentation of breast tissues.
- The Haar wavelet pooling facilitated effective feature extraction and reduced information loss during the network's subsampling stages.
Main Results:
- The proposed U-Net model achieved a mean Intersection over Union (mIOU) of 87.48% for breast tissue segmentation.
- The network demonstrated robust performance and resilience to overfitting.
- High-accuracy segmentation of breast tissues with diverse elastic moduli was achieved, enabling reliable shape reconstruction.
Conclusions:
- The developed U-Net based multi-class semantic segmentation method accurately segments breast tissues, even with varying elastic properties.
- This approach enables the reconstruction of natural breast shapes from supine MRI data.
- The findings offer a promising advancement for applications requiring precise breast morphology modeling.
