Quantitative evaluation of denoising techniques of lung computed tomography images: an experimental investigation

Appropriate selection of denoising method is critical component of lung computed tomography (CT)-based computer aided diagnosis (CAD) systems since noises and artefacts may deteriorate the image quality significantly thereby leading to incorrect diagnosis. This study presents a comparative investigation of various techniques used for denoising lung CT images. Current practices, evaluation measures, research gaps and future challenges in this area are also discussed. Experiments on 20 real-time lung CT images indicate that Gaussian filter with 3 x 3 window size outperformed others achieving high picture signal-to-noise ratio (PSNR), Pratt's figure of merit (PFOM), signal-to-noise ratio (SNR) and root mean square error (RMSE) of 45.476, 97.964, 32.811, 0.948 and 0.008, respectively. Further, this approach also demonstrates good edge retrieval efficiency. Future work is needed to evaluate various filters in clinical practice along with segmentation, feature extraction, and classification of lung nodules in CT images.