Fig 1.
Block diagram of the proposed brain tumor gradation method.
Fig 2.
Detail wavelet subbands of two simulated images having different textures.
Fig 3.
Detail wavelet subbands of two T1-weighted brain MR images with high grade (top) and low grade (bottom) tumors.
Fig 4.
Histograms of co-occurrence of wavelet coefficients for different subbands (top row: Horizontal, middle row: Vertical, and bottom row: Diagonal) of top image of Fig 2.
Fig 5.
Histograms of co-occurrence of wavelet coefficients for different subbands (top row: Horizontal, middle row: Vertical, and bottom row: Diagonal) of top image of Fig 3.
Fig 6.
Histograms of co-occurrence of wavelet coefficients for different subbands (top row: Horizontal, middle row: Vertical, and bottom row: Diagonal) of bottom image of Fig 3.
Table 1.
Textural features obtained from individual subband co-occurrence and multispectral co-occurrence for two example images.
Table 2.
Textural features obtained from individual subband co-occurrence and multispectral co-occurrence for brain images with high and low grade tumors.
Fig 7.
Comparative performance analysis of different decomposition levels of wavelet analysis using ten-fold cross-validation.
Fig 8.
ROC curve obtained using leave-one-out cross-validation for different decomposition levels of wavelet analysis.
Fig 9.
Comparative performance analysis of the proposed fusion method over individual sequences for ten-fold cross-validation.
Fig 10.
ROC curve obtained using leave-one-out cross-validation for the proposed fusion method and individual sequences.
Fig 11.
Comparative performance analysis of different fusion approaches using ten-fold cross-validation.
Fig 12.
ROC curve obtained using leave-one-out cross-validation for different fusion approaches.
Fig 13.
Comparative performance analysis of the proposed multispectral co-occurrence matrix and multi-scale GLCM based methods using ten-fold cross-validation.
Fig 14.
ROC curve obtained using leave-one-out cross-validation for the proposed multispectral co-occurrence matrix and multi-scale GLCM based methods.
Table 3.
Performance analysis of different algorithms using leave-one-out cross-validation.
Table 4.
Accuracy and AUC of different algorithms for ten-fold cross-validation.
Table 5.
Sensitivity and specificity of different algorithms for ten-fold cross-validation.
Table 6.
PPV and NPV of different algorithms for ten-fold cross-validation.
Fig 15.
Calculation of multispectral co-occurrence matrix for all detail wavelet subbands at four directions.
Fig 16.
Illustration of the displacement vector in 3-D space.
Table 7.
Displacement vector for multispectral co-occurrence matrices.