Artificial bee colony clustering with self-adaptive crossover and stepwise search for brain functional parcellation in fMRI data

The emergence of functional magnetic resonance imaging (fMRI) provides a good opportunity for brain functional parcellation. However, the high dimension and low signal-to-noise ratio of fMRI data brings difficulties to the existing parcellation methods. To address the issue, this paper presents a novel brain functional parcellation method based on artificial bee colony clustering (ABCC) algorithm with self-adaptive crossover and stepwise search (called CSABCC). In CSABCC, the preprocessed fMRI data is first mapped into a low-dimensional space by spectral mapping to reduce its dimension and each food source position is encoded as a clustering solution composed of cluster centers. Then, CSABCC utilizes an improved artificial bee colony search procedure with some robustness advantage to seek better food sources, where a self-adaptive crossover is employed to enhance information exchange between individuals and onlooker bees adopt a stepwise search to improve its search capability. Finally, a functional parcellation result is obtained by mapping cluster labels onto the corresponding voxels. The experiments on simulated fMRI data show that CSABCC can generate the parcellation closest to the real result, and these results on real insula fMRI data also demonstrate that CSABCC has better search capability and can produce parcellation structures with stronger functional consistency and regional continuity compared to some other typical algorithms. Moreover, the correctness of the parcellation results is also validated by functional connectivity fingerprints of the corresponding subregions.