TY - GEN
T1 - Multi-Label MambaOut for Quality Assessment of Low-Field Pediatric Brain MR Images
AU - Zhu, Yueyue
AU - Jiang, Haotian
AU - Cai, Rongqing
AU - Chen, Geng
N1 - Publisher Copyright:
© The Author(s) 2025.
PY - 2025
Y1 - 2025
N2 - Magnetic Resonance Imaging (MRI) can be utilized to study the structure of pediatric brains non-invasively. In practice, low-field MRI scanners are widely adopted for pediatric brain imaging. However, the corresponding acquired MRI data usually suffers from severe artifacts, such as noise and motion. Therefore, an effective Quality Assessment (QA) method is essential. To this end, we design a Multi-Label MambaOut (MLMambaOut) model for the low-field pediatric brain MRI QA challenge. Specifically, we view this challenge as a multi-label classification task, utilizing four stages of gated convolution neural network blocks and ML-Decoder to finish the classification with class balance loss. Furthermore, we explore the performance of Mamba and some advanced models for this challenge. We performed extensive experiments on the challenge data, which is low-field and corrupted with seven kinds of artifacts. The results show that our MLMambaOut achieves superior classification results compared with other methods.
AB - Magnetic Resonance Imaging (MRI) can be utilized to study the structure of pediatric brains non-invasively. In practice, low-field MRI scanners are widely adopted for pediatric brain imaging. However, the corresponding acquired MRI data usually suffers from severe artifacts, such as noise and motion. Therefore, an effective Quality Assessment (QA) method is essential. To this end, we design a Multi-Label MambaOut (MLMambaOut) model for the low-field pediatric brain MRI QA challenge. Specifically, we view this challenge as a multi-label classification task, utilizing four stages of gated convolution neural network blocks and ML-Decoder to finish the classification with class balance loss. Furthermore, we explore the performance of Mamba and some advanced models for this challenge. We performed extensive experiments on the challenge data, which is low-field and corrupted with seven kinds of artifacts. The results show that our MLMambaOut achieves superior classification results compared with other methods.
KW - Low-field pediatric brain
KW - Multi-label classification
KW - Quality assessment
UR - https://www.scopus.com/pages/publications/86000449653
U2 - 10.1007/978-3-031-83008-2_1
DO - 10.1007/978-3-031-83008-2_1
M3 - 会议稿件
AN - SCOPUS:86000449653
SN - 9783031830105
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 3
EP - 11
BT - Low Field Pediatric Brain Magnetic Resonance Image Segmentation and Quality Assurance - 1st MICCAI Challenge, LISA 2024, Held in Conjunction with MICCAI 2024, Proceedings
A2 - Lepore, Natasha
A2 - Linguraru, Marius George
PB - Springer Science and Business Media Deutschland GmbH
T2 - 1st MICCAI Challenge on Low Field Pediatric Brain Magnetic Resonance Image Segmentation and Quality Assurance, LISA 2024, held in Conjunction with Medical Image Computing and Computer Assisted Intervention Conference, MICCAI 2024
Y2 - 10 October 2024 through 10 October 2024
ER -