Low-dose computed tomography image denoising using graph wavelet transform with optimal base

Abstract

Noise in electronic components of computed tomography (CT) detectors behaves like a virus that infects visual quality of CT scans and might distort clinical diagnosis. Modern CT detector technology incorporates high-quality electronic components in conjunction with signal and image processing to ensure optimal image quality while retaining benign doses of x-rays. In this study, a new strategy in signal and image processing directions is proposed by finding the most optimal wavelet base for denoising low-dose CT scan data. The process begins by selecting the appropriate wavelet bases for CT image denoising, followed by a wavelet decomposition, thresholding, and reconstruction. Other methods, such as graph wavelet and learning-based, are used to assess the consistency of the outcomes. The wavelet base of biorthogonal 5.5 achieves the highest optimum performance for CT image denoising. Meanwhile, the Daubechies wavelet base is inconsistent and performs poorly compared to the optimum base. This research highlights the importance of wavelet properties such as orthogonality, regularity, and the number of vanishing moments in selecting an appropriate wavelet basis for noise reduction in CT images.