Research On Cerebral CT Image Registration By Adaptive Particle Swarm Optimization

Medical Image Registration is widely used in Digital Medical Treatment. It is very important in Three-dimensional reconstruction, Multi-source medical image fusion, lesions feature extraction and any other diagnosis assistant technology. Accuracy is highly demanded in cerebral diagnosis, and cerebral image registration is the key technology that decide the accuracy and velocity.The common methods that be used to do the medical image registration is Maximization of Mutual Information with optimization methods. Because of the large computation, the capability of the optimization methods can affect the accuracy and velocity of the registration results. The better way to enhance the accuracy and velocity is to use an efficient optimization method combined with the feature of the cerebral images.The Particle Swarm Optimization is a new optimization method advanced these years. This article achieved the latest Adaptive Particle Swarm Optimization, and analyzed how the accuracy and velocity be affected by the dynamic initial parameters and problem space in the problem of brain CT image registration. The result is that the optimal solution increase gradually as a log curve in interval 2 to 20 of the number of particles, and will not increase any more in the interval 20 to infinity. But the time that which the Adaptive Particle Swarm Optimization used is on the rise in the interval 2 to infinity of the number of particles. This shows that the accuracy and velocity could be increased by controlling the initial parameters dynamically.This article use the Adaptive Particle Swarm Optimization combined with Maximization of Mutual Information archived the registration of the brain CT image sequence, and proposed a method which be used in controlling the initial parameters dynamically based on the probability distribution of the black pixels and trend of the initial parameters to keep the accuracy and improve the velocity.As the method is only verified in the registration of the brain CT image sequence, weather it can make the same effect in the CT image of the other organs also need deep exploration.