Automatic brain tumor extraction using fuzzy information fusion

Weibei Dou, Qingmin Liao, Su Ruan, D. Bloyet, Jean-Mac Constans, Yanping ChenProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2002 · DOI

Abstract

This paper presents a fuzzy information fusion method to automatically extract tumor areas of the human brain from multispectral magnetic resonance (MR) images. The multispectral images consist of T1-weighted (T1), proton density (PD), and T2-weighted (T2) feature images, in which signal intensities of a tumor vary across modalities. Because certain tissues are more visible in specific image types, the fusion of information is necessary for accurate segmentation. Our method models fuzzy information about the tumor using membership functions based on a priori radiological knowledge and MR signal characteristics. Experimental results involving five patients demonstrate a mean false-negative rate of 2% and a mean false-positive rate of 1.3% compared to manual expert tracing.

Introduction

Accurately segmenting brain tumors is a critical prerequisite for diagnosis, surgical planning, and treatment monitoring. However, manual segmentation is time-consuming and susceptible to human error or inter-observer subjectivity. The technical challenge lies in the multispectral nature of MRI scans: different sequences (T1, T2, and PD) highlight distinct tissue characteristics. A tumor might appear as a clear "bright" spot in a T2 scan but remain ambiguous in a T1 scan. Relying on a single modality often leads to incomplete or inaccurate extractions. This study proposes a method to "fuse" information from multiple multispectral images to create a more robust, automated boundary for tumor extraction.

Methodology

The proposed pipeline is centered on Fuzzy Information Fusion and consists of four primary stages:

  1. Registration: Since T1, T2, and PD images may be captured at different resolutions or angles, the researchers use the FLIRT tool to align the T2 and PD volumes into the same coordinate space as the T1 image.
  2. Initial Segmentation: The T1 image is segmented into 11 classes using Fuzzy Markovian Random Fields. These classes represent pure tissues (white matter, gray matter, CSF) and "mixclasses" that account for the partial volume effect, where a single pixel contains multiple tissue types.
  3. Fuzzy Modeling: This is the core innovation. Qualitative descriptions from radiologists are translated into quantitative membership functions ($mf$). For instance, if experts define a tumor as "very bright" on T2, a mathematical curve is created where high intensity corresponds to a high "membership" score for being a tumor.
  4. Information Fusion: The system combines scores from all three modalities using a "min" operator (intersection). A region is only classified as a tumor if it scores highly across T1, T2, and PD criteria simultaneously. This helps filter out "false alarms" like edema or CSF that might mimic a tumor in only one modality.
Membership functions for T1, T2, and PD modalities used to model tumor characteristics.

Results and Discussion

The automated system was benchmarked against manual expert segmentation using a clinical dataset of five patient cases. The method achieved a mean false-negative rate of 2% and a mean false-positive rate of 1.3%.

A representative multispectral MR image showing the tumor region targeted for automated extraction.

Visually, the fusion method successfully distinguished the tumor from surrounding edema (swelling), which is a common point of failure for single-modality automated systems. By utilizing the intersection of fuzzy sets, the system ensures that the final extraction represents the area where all three data sources agree, significantly reducing false positives.

Conclusion

This research demonstrates that a knowledge-driven fuzzy fusion approach can achieve high accuracy in brain tumor extraction. By mathematically modeling radiological expertise and handling the inherent ambiguity of tissue boundaries through fuzzy logic, the system provides a reliable alternative to manual tracing. Future work will focus on expanding the dataset to include more diverse tumor types and developing automatic threshold selection to create a truly "hands-off" clinical tool.

Summary updated

Abstract

This paper presents a fuzzy information fusion method to automatically extract tumor areas of human brain from multispectral magnetic resonance (MR) images. The multispectral images consist of T1 -weighted (T1), proton density (PD), and 12-weighted (T2) feature images, in which signal intensities of a tumor are different. Some tissue is more visible in one image type than the others. The fusion of information is therefore necessary. Our method, based on the fusion of information, model the fuzzy information about the tumor by membership functions. Thismodelisation is based on the a priori knowledge of radiology experts and the MR signals of the brain tissues. Three membership functions related to the three images types are proposed according to their characteristics. The brain extraction is then carried out by using the fusion of all three fuzzy information. The experimental results (based on 5 patients studied) show a mean false-negative of 2% and a mean false-positive of 1 .3%, comparing to the results obtained by a radiology using manual tracing.
BibTeX
@inproceedings{dou2002automatic,
  author = {Weibei Dou and Qingmin Liao and Su Ruan and D. Bloyet and Jean-Mac Constans and Yanping Chen},
  title = {{Automatic brain tumor extraction using fuzzy information fusion}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2002},
  doi = {10.1117/12.477203},
}
Figure