Related Experiment Videos
Automatic segmentation of different-sized white matter lesions by voxel probability estimation
Petronella Anbeek1, Koen L Vincken, Matthias J P van Osch
1Department of Radiology, Image Sciences Institute, University Medical Center Utrecht, Heidelberglaan 100, rm E01.335, 3584 CX Utrecht, The Netherlands. nelly@isi.uu.nl
Medical Image Analysis
|September 29, 2004
Summary
This study introduces an automated method for segmenting white matter lesions (WMLs) using K-Nearest Neighbor classification on MRI scans. The algorithm achieves high accuracy, comparable to existing methods, and is suitable for large population studies.
Area of Science:
- Medical Imaging
- Neuroscience
- Computer Vision
Background:
- White matter lesions (WMLs) are common in neurological conditions like multiple sclerosis.
- Accurate segmentation of WMLs is crucial for diagnosis and monitoring disease progression.
- Existing segmentation methods can be time-consuming and may lack robustness.
Purpose of the Study:
- To develop and evaluate a fully automated method for segmenting WMLs on cranial MR imaging.
- To assess the accuracy and reliability of the proposed automated segmentation technique.
- To determine the suitability of the method for large-scale and longitudinal studies.
Main Methods:
- Utilized a K-Nearest Neighbor (KNN) classification technique incorporating voxel intensity and spatial information.
- Generated probability maps per voxel for WML identification.
- Employed thresholding on probability maps to produce binary segmentations.
- Evaluated performance using Receiver Operating Characteristic (ROC) curves and similarity measures (SI, OF, EF).
Main Results:
- The automated segmentation achieved high sensitivity and specificity.
- The method performed well for WMLs larger than 2 cc, with strong correlations between lesion volume and similarity measures (SI, PSI, PEF).
- Accuracy was comparable to existing multiple sclerosis lesion segmentation methods.
Conclusions:
- The developed automated WML segmentation method is accurate and reliable.
- The technique is robust across various lesion sizes and shapes.
- This automated approach is well-suited for WML detection in large population and longitudinal studies.