Fast 3D Brain Multi-thresholding Segmentation Based on Immuno-genetic Algorithm

To solve the problem, that is, the large time-consumption of the complete search (CS), the instability and inaccurateness of the simple genetic algorithm (SGA) and the traditional immuno-genetic algorithm (IGA), a novel 3D brain data segmentation procedure utilizing optimal entropy multi-thresholding method is proposed, in which global maximum entropy for the segmentation is yielded fast by our improved immuno-genetic algorithm (IIGA). Different from the IGA that uses the two-individual mean information entropy for the immune selection, the IIGA assigns the different weight to each term of the total information entropy at the same loci in a pair of individuals, which constructs a better selection scheme and ensures more various individuals to be selected for preserving the diversity of the population. Meanwhile the general expressing form of this kind of selection probability is given. The proposed method also includes the elitist strategy and the adaptive crossover and mutation mechanism to enhance the convergence. Results on 50 simulations demonstrate the real 3D brain volume can be classified to three parts successfully: the white matter, the gray matter and the cerebrospinal fluid on the IDL platform. The stability, accuracy, and speed of our algorithm, compared with other methods, are all improved according to their performance contrasts.