Research And Implement Of Lung CT Images Enhancement Algorithm

Nowadays, medical images have become important assistant means for clinic diagnosis and therapy. CT technology has been widely used in clinics for its high resolution and little harm to human beings. However, degradation of CT medical image can worsen the results of analysis and diagnosis, such as blurring details of the anatomical structure and low contrast during radiographic imaging caused by complexity of body tissue and the limitation of imaging system. At the same time, to improve the contrast of the image from its origin often means longer inspection time or larger radiation dose, and this may cost the patient more money or bring about more risk of overdosed radiation. In order to make the doctor observe and diagnose the anatomy structure and pathological part of the patient more effectively, it has important clinical values in CT medical image enhancement pre-processing.In this thesis, we firstly make the analysis and summarization of medical image enhancement techniques in existence, and then mainly discuss the image enhancement algorithms based on histogram analysis and the image enhancement technique based on the Retinex theory. We put forwards correspondence methods to solve the concrete problems met in the course of specific pulmonary CT image enhancement. The main achievements are summarized as following:1) Connecting with the characteristics of lung CT image such as wide dynamic range, abundant details, relatively small reagions of interest and poor contrast between ROI and the surrounding areapoor contrast, meanwhile, directing against the defects of the related image enhancement algorithms in existence based on histogram processing, we have developed an improved contrast enhancement algorithm based on local feature analysis for lung CT images. The method performs a local modified contrast stretching using a non-linear transfer function based on the analysis of image characteristics, and makes gray scale new mappings adaptively. The experimental results based on MATLAB platform, show that the proposed method effectively overcomes the shortcomings of histogram equalization such as over-enhancement, level saturation effects, also the defects of local or adaptive methods including larger computational burden, weak noise suppression ability etc.; it improves the global visual and enhances the detail anatomic structure effectively. Now, the method is used as a pre-processing procedure in a Lung Cancer CAD system.2) In view of the unique features of pulmonary CT image and aiming at the disadvantages of the center/surround Retinex related image enhancement algorithms in existence, we developed an improved image enhancement method for lung CT image based on single-scale Retinex. In the proposed method, we make the reflectance component enhancement in a nonlinear way and compensate the approximate illumination image estimation, and obtain the enhanced image restoration with a gamma correction function. Our method is implemented in MATLAB and the experimental results show that the proposed method reduces the computational complexity, avoids the halo artifacts in conventional SSR and can simultaneously provide dynamic range compression, sharpening and color constancy. The algorithm yields better performance of lung CT image enhancement over the general method and is proved as a useful method which can meet the requirements for clinical diagnoses applications and image preprocessing part of the computer-assisted diagnosis system.