Brain Disease Classification Using Contourlet Transform and GLCM Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying brain diseases using MRI images face challenges in capturing curve-like features and directional selectivity, with existing wavelet transforms like DWT and SWT being limited in their ability to manage 2D-singularities and directional selectivity.
Innovation Solution
The proposed method employs a pyramidal directional filter bank contourlet transform (PDFB-CT) combined with gray-level co-occurrence matrix (GLCM) texture features and probabilistic principal component analysis (PPCA) for feature extraction and selection, followed by multi-kernel support vector machine (MK-SVM) classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If wavelet transform (DWT/SWT) is used for feature extraction from MRI brain images, then feature extraction capability is provided, but directional selectivity and curve-like feature capture are insufficient
Solution Approach 1:
The patent segments the frequency spectrum into multiple directional bands using a directional filter bank structure. The contourlet transform decomposes the image into approximation and detail coefficients at multiple scales and orientations, with each directional band capturing specific orientation information. This segmentation enables selective analysis of features in different directions, resolving the limitation of wavelet transforms that cannot effectively capture curve-like structures with specific orientations.
Solution Approach 2:
The patent introduces an additional dimensional aspect by incorporating directional filtering in the frequency domain. Instead of only multi-resolution decomposition like wavelets, the contourlet transform adds directional selectivity as a new dimension of analysis. The directional filter bank creates multiple sub-bands that capture features at different orientations, effectively adding a directional dimension to the feature extraction process and enabling better capture of curve-like structures.
2Ease of manufacture
If traditional wavelet transform is used, then computational simplicity is maintained, but ability to manage 2D-singularities and directional features is limited
Solution Approach 1:
The patent segments the 2D frequency spectrum into multiple directional sectors using a radial symmetry structure. The directional filter bank divides the frequency plane into wedge-shaped regions, each capturing singularities in specific directions. This segmentation allows the system to detect and analyze 2D-singularities (edges, curves, corners) with directional specificity, overcoming the limitation of traditional wavelets that treat all directions uniformly and cannot effectively capture the geometric structure of 2D-singularities.
3Measurement precision
If more comprehensive feature extraction is performed to improve classification accuracy, then dimensionality of feature vector increases, but classification efficiency decreases
Solution Approach 1:
The patent extracts only the most relevant features by selecting specific directional sub-bands and approximation coefficients that contain the most diagnostic information for brain disease classification. Instead of using all possible wavelet coefficients, the contourlet transform allows selective extraction of features from specific directional bands, removing redundant information and reducing feature dimensionality while maintaining classification accuracy. This selective extraction improves computational efficiency by reducing the number of features that need to be processed by the classifier.
Data Source
AI summary
The present invention relates to a method of providing diagnostic information for brain diseases classification, which can classify brain diseases in an improved and automated manner through magnetic resonance image pre-processing, steps of contourlet transform, steps of feature extraction and selection, and steps of cross-validation. The present invention relates to a diagnostic information providing method capable of providing an optimal diagnostic means. The present invention relates to a method for providing diagnostic information for brain diseases classification, and relates to a method for providing an optimal diagnostic means for classifying brain diseases in an improved and automated manner through the steps of the magnetic resonance imaging pre-processing, contourlet transform, feature extraction and selection, and cross-validation.