Atlas Guided Identification of Brain Structures by Combining 3D Segmentation and SVM Classification

Ayelet Akselrod-Ballin, Meirav Galun, Moshe John Gomori, Ronen Basri, Achi Brandt

Abstract

Automated segmentation of brain structures in Magnetic Resonance Imaging (MRI) is a critical yet challenging task due to artifacts such as intensity non-uniformity and partial volume effects. This paper proposes a hybrid framework that integrates a fast, multiscale 3D segmentation algorithm—Segmentation by Weighted Aggregation (SWA)—with a Support Vector Machine (SVM) classifier. By incorporating spatial prior knowledge from a probabilistic brain atlas, the system achieves state-of-the-art accuracy in identifying white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Evaluation on the IBSR dataset demonstrates the method's robustness against noise and its superiority over existing low-level and statistical segmentation techniques.

Introduction

Accurate identification of brain tissues in MRI is essential for neurological diagnosis and therapy planning. However, automated methods must overcome significant hurdles, including Intensity Non-uniformity (INU), where magnetic field inconsistencies cause tissue types to appear varied across the scan, and Partial Volume Effects (PVE), where voxels contain multiple tissue types. Traditional low-level algorithms like k-means often lack global context, while complex deformable models can be computationally prohibitive. This research seeks a middle ground: a fast, hierarchical method informed by global anatomical priors.

Methodology

The proposed pipeline consists of three primary stages:

  1. Multiscale Segmentation (SWA): Utilizing "Segmentation by Weighted Aggregation," the method recursively groups similar adjacent voxels into larger aggregates, forming a hierarchical pyramid. This process maintains linear time complexity, ensuring high computational efficiency.
  2. Feature Extraction: For each segment in the pyramid, a diverse set of features is extracted, including intensity statistics (mean, variance), geometric properties (via PCA), and atlas-based spatial priors. The scans are aligned to the ICBM452 probabilistic atlas to calculate the likelihood of a segment belonging to WM, GM, or CSF based on its location.
  3. SVM Classification: These multiscale features are fed into an SVM classifier with a Radial Basis Function (RBF) kernel. The SVM learns to classify segments based on the rich contextual information provided by the hierarchical structure and anatomical priors.

Experimental Setup

The method was validated using the Internet Brain Segmentation Repository (IBSR), which contains 20 coronal T1-weighted real MRI scans of normal subjects. These scans include expert-provided manual segmentations as ground truth. The dataset is particularly useful for testing robustness, as it includes varying levels of signal-to-noise ratios and intensity inhomogeneities.

Results

The proposed method was compared against ten established algorithms using the Jaccard similarity (J-score). The results indicate that the hybrid SVM-SWA approach consistently outperforms existing methods, particularly in challenging tissue classes like CSF.

Visual results of the proposed segmentation method showing 2D slices and 3D reconstructions of brain tissues.

As shown in the quantitative analysis, the method achieved a J-score of 0.680 for Gray Matter and 0.669 for White Matter. Notably, the CSF segmentation reached a J-score of 0.346, a significant improvement over many existing methods that typically score below 0.1 for this complex structure.

Table of performance metrics across different tissue classes, including Overlap, False Positive (FP), Kappa (κ), and Jaccard similarity (J).

Discussion

A key strength of this approach is its hierarchical nature; by analyzing voxels within larger neighborhoods, the algorithm mimics the multiscale perspective of human radiologists. Furthermore, the method demonstrates remarkable robustness to Intensity Non-uniformity without requiring explicit bias correction steps. By treating the probabilistic atlas as a guide rather than a rigid template, the SVM maintains the flexibility necessary to handle individual anatomical variations.

Conclusion

The integration of multiscale segmentation with SVM classification and atlas priors provides a robust and efficient solution for brain MRI segmentation. While the current study focused on healthy subjects, future work aims to extend this framework to pathological cases, such as brains with tumors or significant structural deformities, and to refine the system for identifying granular internal substructures like the thalamus or hippocampus.

Summary updated

Abstract

This study presents a novel automatic approach for the identification of anatomical brain structures in magnetic resonance images (MRI). The method combines a fast multiscale multi-channel three dimensional (3D) segmentation algorithm providing a rich feature vocabulary together with a support vector machine (SVM) based classifier. The segmentation produces a full hierarchy of segments, expressed by an irregular pyramid with only linear time complexity. The pyramid provides a rich, adaptive representation of the image, enabling detection of various anatomical structures at different scales. A key aspect of the approach is the thorough set of multiscale measures employed throughout the segmentation process which are also provided at its end for clinical analysis. These features include in particular the prior probability knowledge of anatomic structures due to the use of an MRI probabilistic atlas. An SVM classifier is trained based on this set of features to identify the brain structures. We validated the approach using a gold standard real brain MRI data set. Comparison of the results with existing algorithms displays the promise of our approach.
BibTeX
@inproceedings{akselrodballin2006atlas,
  author = {Ayelet Akselrod-Ballin and Meirav Galun and Moshe John Gomori and Ronen Basri and Achi Brandt},
  title = {{Atlas Guided Identification of Brain Structures by Combining 3D Segmentation and SVM Classification}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2006},
  doi = {10.1007/11866763\_26},
}
Figure