Related Experiment Video
Updated: Apr 8, 2026
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
49.7K
Rotation-invariant multi-contrast non-local means for MS lesion segmentation
Nicolas Guizard1, Pierrick Coupé2, Vladimir S Fonov1
1Montreal Neurological Institute, McGill University, Canada.
Neuroimage. Clinical
|June 25, 2015
Summary
This study introduces a novel method for segmenting multiple sclerosis (MS) lesions in brain MRIs. The rotation-invariant multi-contrast non-local means segmentation (RMNMS) method accurately identifies MS lesions of various sizes and orientations.
Area of Science:
- Medical Imaging
- Neuroscience
- Computer Vision
Background:
- Accurate multiple sclerosis (MS) lesion segmentation is vital for disease management and treatment evaluation.
- Lesion variability in size, location, and intensity presents significant challenges for automated segmentation methods.
- Existing automated methods struggle with the diverse characteristics of MS lesions in 3D magnetic resonance (MR) images.
Purpose of the Study:
- To develop and validate a novel supervised method for accurate MS lesion segmentation from 3D MR images.
- To improve the robustness and accuracy of automated MS lesion detection, addressing variability in lesion appearance.
- To introduce the rotation-invariant multi-contrast non-local means segmentation (RMNMS) method for enhanced MS lesion identification.
Main Methods:
- Proposed a supervised segmentation approach utilizing non-local means (NLM) for 3D MR images.
- Developed a multi-channel, rotation-invariant distance measure to handle diverse MS lesion characteristics.
- Implemented the rotation-invariant multi-contrast non-local means segmentation (RMNMS) algorithm for lesion segmentation.
Main Results:
- Internal validation showed good performance with Dice similarity of 60.1% and sensitivity of 75.4%.
- Strong correlation (R²=0.91) observed between expert and automatic lesion load volumes.
- Demonstrated robust lesion detection across different sizes and locations (lesion detection rate=79.8%) on an independent dataset.
- Achieved competitive results compared to state-of-the-art methods on the MS Grand Challenge (MSGC) dataset.
Conclusions:
- The RMNMS method accurately and robustly segments multiple sclerosis lesions from 3D MR images.
- The proposed method demonstrates effectiveness in capturing lesion spatial distribution irrespective of orientation, shape, or size.
- RMNMS offers a promising tool for clinical evaluation of disease burden and treatment efficacy in multiple sclerosis.
