Effective immune genetic algorithm for segmentation of 3D brain images

To solve large time-consumption of the complete search (CS), and the instability and inaccurateness of the simple genetic algorithm (SGA), an effective 3D brain images segmentation procedure, utilizing optimal entropy multi-thresholding method, was proposed. Global maximum entropy for the segmentation was yielded fast by the combination of the immune genetic algorithm (IGA) and simulated annealing (SA). Compared to the SGA, the IGA constructs a better selection scheme and ensures various individuals to be selected for preserving the diversity of the population. Meanwhile, the optimal entropy function of 3D medical images is stretched by the SA to construct the new fitness function, and the general expressing form of the selection probability for IGA is also given. Furthermore, to enhance the convergence of our algorithm, the proposed method includes the elitist strategy and the adaptive crossover and mutation mechanism. Results of 100 simulations demonstrate that the 3D brain volume can be successfully classified into three parts: the white matter, the gray matter and the cerebrospinal fluid on the IDL platform. The stability and accuracy of the algorithm, compared with the SGA and IGA, are all improved according to their performance contrasts.