Research On Low-Dose CT Image Reconstruction Based On Random Forest

With the continuous development and extensive application of CT imaging technology,low-dose CT image reconstruction research as an important branch of modern medical imaging has also been greatly developed.In order to reduce the radiation dose generated during CT scanning and at the same time ensure that the quality of CT images will not be reduced,the study of low-dose CT image reconstruction has great value.The traditional reconstruction algorithm is very easy to cause the phenomenon of under-fitting and over-fitting,which makes the reconstructed image fuzzy,and the ability to recover some high-frequency information of the original image is also insufficient.The algorithm studied in this paper adopts the mode of coupled dictionary learning to bring the idea of random forest algorithm into the training process,which can better reconstruct high-quality CT images and reduce the training and reconstruction time of the algorithm,which greatly improves the algorithm effectiveness.The main research work of this article is as follows:(1)On the basis of traditional dictionary learning,a super-resolution reconstruction algorithm based on coupled dictionary learning based on non-local similarity constraints is proposed.This algorithm transforms dictionary training in traditional algorithms into a mapping between training dictionary and sparse representation.Relationship;in the reconstruction process,first initialize and determine two of the variables to train and update the other,so as to repeatedly obtain the convergence value of a regression function,that is,the optimal solution of the variable;and finally add non-local similarity constraints to This strengthens the robustness of the algorithm in suppressing noise and retaining the original image edge or high-frequency information.(2)Take advantage of the characteristics of random forest strong classifier and prediction model to combine it with coupled dictionary learning.In the training phase,the original dictionary learning training process is replaced with a random forest training process,and a model that can accurately model the mapping relationship between images is found through the characteristics of the random forest;in the reconstruction stage,this mapping relationship model and the input image Combine to obtain the optimal mapping relationship,and use it to replace the mapping relationship obtained by the previous dictionary training into the coupled dictionary learning for calculation to obtain the reconstructed image under the random forest prediction model,which greatly reduces the training time And reconstruction time,improve the efficiency of the algorithm.(3)Design a controlled experiment to verify that the super-resolution reconstruction algorithm based on coupled dictionary learning in this paper has stronger capabilities in suppressing noise and retaining original image information than traditional dictionary learning algorithms;the algorithm after combining random forest and coupled dictionary learning It further improves the noise reduction ability of the original coupled dictionary learning,and greatly reduces the training time of the algorithm and saves the calculation cost.