Coordinate Transformations Make Segmentation Models More Data-Efficient

doi:10.1007/978-3-032-14417-1_7

2026-01-01

Mahbod Issaiy *

DOI：10.1007/978-3-032-14417-1_7

原文链接

原文求助

摘要

En 中文

超低场（0.064T）磁共振成像（MRI）系统实现了便携式脑部成像，但由于信噪比较低和分辨率有限，对自动化分割构成了重大挑战。我们提出了一种基于坐标变换的深度学习方法，用于儿童脑结构分割，通过球面和椭球面坐标映射解析处理几何变异性。我们的方法采用在笛卡尔、球面和椭球面空间中训练的SwinUNETR模型集成，并结合两种新型损失函数：通过二维正交投影进行形状感知监督的投影Dice损失，以及使用坐标适当距离度量的坐标感知软豪斯多夫损失。在LISA25挑战数据集上进行评估，我们的集成方法在 hippocampus 分割上实现了Dice系数0.72±0.17，在 basal ganglia 分割上实现了0.85±0.05的竞争性性能。虽然坐标变换提供了规范的几何处理，但逆变换伪影限制了其个体有效性。新型损失函数在医学图像分割中显示出明显优势，提升了资源受限环境下便携式脑部MRI系统的自动化分析能力。源代码可在 https://github.com/mahbodez/LISA25-public 获得。

Keyword:

Brain MRI segmentation

Low-field MRI

Coordinate transformations

Deep learning

Pediatric neuroimaging

LISA25 challenge

期刊

LOW FIELD PEDIATRIC BRAIN MAGNETIC RESONANCE IMAGE SEGMENTATION AND QUALITY ASSURANCE, LISA 2025

IF:

论文数:

被引数:

机构

tehran university of medical sciences

学者数:

3.0W

论文数: 1.8W

被引数: 30

引用论文

Nonlocal Transform-Domain Filter for Volumetric Data Denoising and Reconstruction

IEEE TRANSACTIONS ON IMAGE PROCESSING

IF13.7

2013-01-01

751

PREAI

Maggioni, Matteo; Katkovnik, Vladimir; Egiazarian, Karen; Foi, Alessandro

Assessment of Brain Injury Using Portable, Low-Field Magnetic Resonance Imaging at the Bedside of Critically Ill Patients

JAMA Neurology

IF0

2021-01-01

OAAI

Kevin N. Sheth; Mercy H. Mazurek; Matthew M. Yuen; Bradley A. Cahn; Jill T. Shah; Adrienne Ward; Jennifer A. Kim; Emily J. Gilmore; Guido J. Falcone; Nils Petersen; Kevin T. Gobeske; Firas Kaddouh; David Y. Hwang; Joseph Schindler; Lauren Sansing; Charles Matouk; Jonathan Rothberg; Gordon Sze; Jonathan Siner; Matthew S. Rosen; Serena Spudich; W. Taylor Kimberly

Two years changes in the development of caudate nucleus are involved in restricted repetitive behaviors in 2–5-year-old children with autism spectrum disorder

Developmental Cognitive Neuroscience

IF0

2016-06-01

OAAI

Ting Qiu; Chen Chang; Yun Li; Lu Qian; Chao Yong Xiao; Ting Xiao; Xiang Xiao; Yun Hua Xiao; Kang Kang Chu; Mark H. Lewis; Xiaoyan Ke

Infant Hippocampal Segmentation in Ultra-Low-Field MRI Using External Datasets with Diverse Field Strengths

Lecture Notes in Computer Science

IF0

2025-01-01

PREAI

Zhou,Weichen; Li,Jingyu; Wang,Xi; Wang,Yuwan; Lyu,Mengye