The explosion operation of fireworks algorithm boosts the coral reef optimization for multimodal medical image registration

Medical image registration is becoming increasingly important in diagnosis and treatment planning, as it enables to align and integrate different images having a shared content, which obtained under different conditions, into a single representation. The most challenging issues of registration is finding a optimal transformation which settled by optimization methods. With this work, we present a Fireworks Algorithm (FWA) boosts the Coral Reefs Optimization Algorithm (CRO) for medical image registration (FWBCR). Firstly, the Coral Reefs Optimization is easily trapping in local optimum, we enhance the exploration ability of the CRO by integrated with a explosion operator of the FWA. At the Broadcast Spawning stage of CRO, a large fraction of corals are applying the crossover operator which has a good exploitation ability and a small fraction of corals with good solutions are applying the explosion operator which has a good exploration ability. Therefore, the algorithm we proposed is well solved the trades off between exploitation and exploration. Secondly, the explosion amplitude affects the search ability. A larger explosion amplitude benefit for the exploration in the early stages of the algorithm. Contrary, a smaller explosion amplitude benefit for the exploitation in the late stages of the algorithm. To deal with these problems, we design an adaptive explosion amplitude which depends on the fitness of the best solution and the worst solution. Thirdly, the phenomenon of the explosion operator is to generate lots of sparks. It is hard to select a small number of sparks with less computation effort while to guarantee the diversity of the swarm for next generation from a large number of sparks. To overcome this difficulty, we exploring the use of the clustering algorithm on the sparks then get some cluster centers as the new solutions. In this way, the diversity of the swarm can be ensured and the computation effort can be greatly reduced through reducing the number of evaluations. Finally, inspired by the Differential Evolution Algorithm, we construct a differential migration vector(DMV) with the most promising right direction and adaptive length, and to generate elite solutions by adding the DMV to the position of the spark centers. The FWBCR has been tested in numerous experiments on benchmark datasets include six kind of different modality images, from up to eighteen different patients, which can make up 54 multimodal registration scenarios. For registration precision, FWBCR obtained best in 19 scenarios, BBO-EL obtained best in 22 scenarios, while CRO-SL obtained best in 13 scenarios, which demonstrated that FWBCR outperformed the CRO variants like CRO-SL in most cases and as good as the state-of-the-art in the registration field. In addition, compared with BBO-EL, FWBCR achieve fast convergence rate and increase computing performance by 30%. Medical image registration is becoming increasingly important in diagnosis and treatment planning, as it enables to align and integrate different images having a shared content, which obtained under different conditions, into a single representation. The most challenging issues of registration is finding a optimal transformation which settled by optimization methods. With this work, we present a Fireworks Algorithm (FWA) boosts the Coral Reefs Optimization Algorithm (CRO) for medical image registration (FWBCR). Firstly, the Coral Reefs Optimization is easily trapping in local optimum, we enhance the exploration ability of the CRO by integrated with a explosion operator of the FWA. At the Broadcast Spawning stage of CRO, a large fraction of corals are applying the crossover operator which has a good exploitation ability and a small fraction of corals with good solutions are applying the explosion operator which has a good exploration ability. Therefore, the algorithm we proposed is well solved the trades off between exploitation and exploration. Secondly, the explosion amplitude affects the search ability. A larger explosion amplitude benefit for the exploration in the early stages of the algorithm. Contrary, a smaller explosion amplitude benefit for the exploitation in the late stages of the algorithm. To deal with these problems, we design an adaptive explosion amplitude which depends on the fitness of the best solution and the worst solution. Thirdly, the phenomenon of the explosion operator is to generate lots of sparks. It is hard to select a small number of sparks with less computation effort while to guarantee the diversity of the swarm for next generation from a large number of sparks. To overcome this difficulty, we exploring the use of the clustering algorithm on the sparks then get some cluster centers as the new solutions. In this way, the diversity of the swarm can be ensured and the computation effort can be greatly reduced through reducing the number of evaluations. Finally, inspired by the Differential Evolution Algorithm, we construct a differential migration vector(DMV) with the most promising right direction and adaptive length, and to generate elite solutions by adding the DMV to the position of the spark centers. The FWBCR has been tested in numerous experiments on benchmark datasets include six kind of different modality images, from up to eighteen different patients, which can make up 54 multimodal registration scenarios. For registration precision, FWBCR obtained best in 19 scenarios, BBO-EL obtained best in 22 scenarios, while CRO-SL obtained best in 13 scenarios, which demonstrated that FWBCR outperformed the CRO variants like CRO-SL in most cases and as good as the state-of-the-art in the registration field. In addition, compared with BBO-EL, FWBCR achieve fast convergence rate and increase computing performance by 30%.

Automated summary

Medical image registration is an increasingly important technique that requires finding the optimal transformation. A new method combining Fireworks Algorithm with Coral Reefs Optimization Algorithm has been proposed in the study and demonstrated superior performance in experiments compared to other algorithms, achieving faster convergence and increased efficiency.