Evaluation of DWT denoise method on X-ray images acquired using flat detector

The aim of this paper is to develop a novel procedure to reduce noise in X-ray images in order to expose patient at the lowest X-ray in general radiography acquisition. The method is based on the analysis of two X-ray images of Pro-Digi phantom acquired using two different X-Ray doses. Acquisition was done using general X-ray machine with flat detector, Multix Fusion (Siemens, Erlangen, Germany). The high dose image is considered as reference-standard image and the lower X-ray dose is used as noisy image. In this paper, seven Regions Of Interest (ROIs) with different contrast were extracted from two X-ray images of Pro-Digi phantom at the same locations. First, we analysis their histograms. Then, we looked for noise classification. For that purpose, ROIs images were filtered using medium and mean filters. The results showed that the medium and mean filters were not able to reduce these noises. So, we have proposed a novel denoising and frequential noise localization procedure by using discrete wavelet transform (DWT) and thresholding methods (TM) which call DWTTM. The low X-ray dose image was decomposed using the DWT and then we used threshold to remove low intensity energy coefficients and we did inverse DWT to reconstruct denoised low dose ROIs image. The originality of the DWTTM is explained by the choice of the approximation image to denoise according to the decomposition level, depending of the noise localization in the ROIs on frequential scale. The choice of the decomposition level and the threshold value are depending of the convergence of the mean value of each of filtered ROIs to the corresponding value of reference image ROIs. To evaluate our results, we use the Contrast to Noise Ratio (CNR) and the Signal Noise Ratio (SNR). Mean values of ROIs regions serve also to calculate the CNR and SNR. The CNR (respectively SNR) values were compared for the both ROI's images (reference image and denoised image) after threshold the approximated sub-images obtained with DWT of low X-ray image. Note that the denoising operation is accomplished separately on each of the ROIs. As results, the SNR and CNR show the existence of noise in the reduced dose image. Also, after the denoising of each ROI of low X_ray image, by threshold the noisy energy coefficients on the approximation DWT sub-image, the evaluation of the results using these descriptors highlights the reduction of the noise.