Parallel adaptive simulated annealing for computer-aided measurement in functional MRI analysis

Simulated annealing (SA) is known to be one of the most efficient heuristic algorithms for complex nonlinear optimization problems. However, it suffers from the large amount of computing time required to obtain a near-optimal solution. To overcome this problem, a parallel version of the algorithm is worthy of evaluation. This report presents a Parallel Adaptive Simulated Annealing (PASA) algorithm for computer-aided measurement in locating the activation area of functional magnetic resonance images (fMRI). The parallel paradigm is based on a coarse-grained (or island) model performed on a cluster of PCs by using a message-passing interface (MPI) for the information interchange. Performance of this approach is evaluated by computing the receiver operating characteristic (ROC) area and Jaccard similarity. Experimental results show that the coarse-grained PASA outperforms other approaches and can also efficiently and consistently extract activities with different contrast-to-noise ratios and activation-area sizes.