Comparative Analysis of Wavelet, Curvelet, and Contourlet Transforms for Denoising Liver CT Images

Abstract

Liver cancer is one of the leading causes of mortality worldwide, and early detection using CT imaging plays a critical role in patient survival. However, the quality of CT scans is often degraded by noise, particularly in low-dose imaging intended to minimize radiation exposure. This study presents a comprehensive comparison of three multiresolution transform techniques, namely the wavelet, curvelet, and contourlet transforms, for denoising liver CT images. The performance of these methods was evaluated using multiple quantitative metrics (PSNR, SNR, and NCC) under varying noise levels, complemented by detailed qualitative assessments of the reconstructed images. Statistical analysis demonstrated that the curvelet transform achieved significantly superior results (p<0.01) compared with wavelet and contourlet methods across all noise variances. For each transform, threshold parameter choices were carefully examined, with a focus onthe BayesShrink thresholding rule applied consistently in all cases. Visual analysis further confirmed that curvelet-based denoising preserved more anatomically relevant structures, including liver regions, vessel details, and tissue textures. These findings underscore the value of curvelet transforms for enhancing image quality in Computer-Aided Diagnosis (CAD) systems for liver pathologies, ensuring diagnostic features are maintained while effectively suppressing noise.