Brain MRIs classification based on 2D SWD-MF-DFA

Background:

To improve imaging classification accuracy, we modify the traditional 2D multifractal trend fluctuation analysis (MF-DFA) method to better preserve local feature values. Inspired by MF-DFA, we develop a novel method for extracting eigenvalues, enhancing the precision of imaging analysis.

New method:

In this paper, we propose an enhanced algorithm building upon the traditional 2D MF-DFA. Our approach introduces a 2D sliding window (SWD) technique for feature value extraction. Initially, the local generalized Hurst index of the imaging is derived using the SWD algorithm, based on MF-DFA principles. Subsequently, the generalized Hurst index is recalculated for the digital matrix formed by these local Hurst indexes. These vectors are then input into a support vector machine (SVM) for classification. This methodology seeks to refine the traditional 2D MF-DFA by more effectively preserving local feature values in imaging.

Results:

The classification accuracy of the SWD eigenvalue extraction method based on 2D MF-DFA reaches 91.54%.

Comparison with existing methods:

We employ brain magnetic resonance imaging (MRI) data sets to evaluate the efficacy of both the conventional 2D MF-DFA method and our proposed feature value extraction technique. Both methods are applied alongside a SVM for classification. The findings reveal that the conventional 2D MF-DFA method yields a classification accuracy of 59.40%, while our SWD feature value extraction method attains a classification accuracy of 91.54%.

Conclusion:

This substantial performance enhancement underscores the superiority of the SWD approach over the conventional method.