A novel MF-DFA-Phase-Field hybrid MRIs classification system

Accurate classification of magnetic resonance imaging (MRI) is an urgent need in clinical medicine. In this study, we explore an integrated classification model using multifractal detrended fluctuation analysis (MF-DFA) and Phase-Field models to develop a novel classifier that ensures high classification accuracy. A nonlinear hyperplane can be generated through the Phase-Field model and a dataset can be subsequently classified. First, two different types of MRI datasets are characterized in two-dimensional (2D) and three-dimensional (3D) spaces after feature extraction using Kolmogorov complexity (KC), Shannon entropy (SNE) and Higuchi's Hurst exponent (HHE). For small samples, a classification effect with 100% Accuracy, Recall and Precision can be achieved. However, for large samples, a good classification effect cannot be achieved. Therefore, we propose a novel MF-DFA-Phase-Field hybrid MRI classification method that also achieves a good classification effect on large samples. The effectiveness and robustness of the proposed MF-DFA-Phase-Field classifier will be analyzed using the generated synthetic data. Subsequently, the two datasets are represented in 2D and 3D computational spaces, where the generalized Hurst exponent computed by MF-DFA is used as the representation coordinate. For the first MRI dataset, the Accuracy, Recall, and Precision of the results for the classification metrics were 100%. In addition, we adopted another dataset with more complex image features and a larger sample size, achieving Accuracy, Recall and Precision of 92.65%, 92.85% and 92.87%, respectively. The Accuracy, Recall and Precision of the classification model based on a support vector machine (SVM) using the same dataset with 11 Hurst exponents as input vectors are 86.32%, 88.50% and 87.46% respectively. These results are all less than those of the proposed model. Similarly, our model performed better in other aspects than those by other scholars, such as MP-CNN and FC-CNN.