Study and summary of Medical Image Registration Based on Itk

Http://blog.sina.com.cn/s/blog_62ccbb810100hven.html

There is not much time to invest in Itk. In the learning process, we mainly understand the Registry framework, and are familiar with various transformations, principles, optimizers, and interpolation. Today, I have made some adjustments to my assignment. I will not focus on my subsequent work. I will summarize my understanding of Itk registry some time ago, it cannot be a waste of time!

   The registration process mainly includes four modules: transformation, interpolation, criterion, and optimization. The entire registration implementation process can be considered as an optimization process. After the transformation method is determined, the image to be registered is transformed to the coordinate system of the reference image. Because the transformed position may not be on the grid point, the brightness value of the transformed point is obtained through interpolation, such as linear interpolation and neighborhood interpolation.

   Metric (criterion), such as mean variance and mutual information criterion. It is used to calculate the differences of the changed image. Its input includes: reference image and image to be registered, transform function and interpolator ). The value calculated by using the criterion will be optimized by the optimizer, and the optimized parameters will return iterations before registration, until the requirements are met.

   The optimizer has two types: 1. singalvalue, that is, the optimizer returns the cost function of a single value, and 2. returns the multi-value, that is, muti-value.

   Registration is based on the above four parts, for different graphics, such as single-mode, multimode, to use different principles, transformations, optimizers and interpolation methods. In this registration framework, different parts are transformed when setting parameters, which is also difficult. Different optimizers often have different parameter settings. Therefore, you need to understand the optimizer and registration algorithm. For example, the Rado transformation has a rotating matrix and the translation volume. Muti-solution registration (Multi-Resolution Registration) must change the step size and so on.

 Set the pixel type and dimension of the image to be registered and the reference image. Instantiate the above four parts and combine them into registry, set parameters for each part, and trigger registration. The registration result is obtained through the final parameter re-sampling with the registration image. Compare the differences between pre-registration and post-registration. Instantiation usually defines -- tyoedef first, then points to the smart pointer, and then calls the Set * () function to set parameters through the pointer.

  During the registration process, * filter is used to process the image based on the actual situation. For example, if you require high accuracy in registration, use cast * filter to convert the image to the float type, and convert the image to Char or Int type storage after registration. If the image noise is obvious, use * filter to remove noise before registration. Some also use filters to retain the edge and perform registration.

  The parameter input of the Itk is input in the property dubug of the project, such as the file name of the registration image, some parameters of the optimizer, the brightness outside the area of the image registration after the change, name of the output image. The specific code is not attached here. As long as you understand the above process, the code looks better.

  After changing the task, we should make new plans for the task and have a near-long-term plan. The biggest competitor is himself. tianxingjian, a gentleman strives for self-improvement, and a gentleman carries things with high morality!