Related Experiment Videos
Statistical analysis of fractal-based brain tumor detection algorithms
Justin M Zook1, Khan M Iftekharuddin
1Department of Biomedical Engineering, The University of Memphis, TN 38152-3810, USA.
Magnetic Resonance Imaging
|July 30, 2005
Summary
Fractal dimension (FD) analysis can detect brain tumors in MR and CT images. This study statistically validates FD algorithms, showing tumor regions have significantly lower FD than healthy tissue.
Area of Science:
- Medical imaging analysis
- Computational geometry
- Radiology
Background:
- Fractals, characterized by noninteger fractal dimensions (FD), have applications in biomedical recognition.
- Previous research demonstrated the utility of FD in brain tumor detection when a reference non-tumor image was available.
- Existing methods for brain tumor detection using FD require a control image for comparison.
Purpose of the Study:
- To statistically validate fractal dimension (FD) analysis for brain tumor detection in a larger dataset of real MR and CT images.
- To develop and evaluate FD techniques that do not require a reference non-tumor image.
- To compare the performance of novel FD algorithms against existing fractal-based methods and manual segmentation.
Main Methods:
- Statistical validation of FD analysis on 80 real MR and CT brain images.
- Development of 'half-image' and 'whole-image' techniques for FD computation, removing the need for a reference image.
- Comparison of developed algorithms with literature-based fractal methods and validation against manually segmented tumor images.
Main Results:
- The tumor region exhibited a statistically significant lower fractal dimension (FD) compared to the non-tumor area across most FD algorithms studied.
- The developed whole-image technique successfully computed tumor FD without requiring a reference non-tumor image.
- Statistical validation confirmed the efficacy of the FD algorithms in differentiating tumorous tissue.
Conclusions:
- Fractal dimension analysis, particularly using the developed algorithms, shows significant promise for the detection and localization of brain tumors in MR and CT images.
- The removal of the need for a reference image enhances the practicality and applicability of FD-based brain tumor detection methods.
- These findings support the exploitation of FD algorithms for automated or semi-automated brain tumor identification in clinical settings.