Noise Suppression of Computed Tomography (CT) Images Using Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)

In this study, an in-house residual encoder-decoder convolutional neural network (RED-CNN)-based algorithm was composed and trained using images of cylindrical polymethyl-methacrylate (PMMA) phantom with a diameter 26 cm at different simulated noise levels. The model tested on 21 × elliptical PMMA computed tomography (CT) to evaluate its denoising capability signal ratio (SNR), comparative peak signal-to-noise (cPSNR), structural similarity (SSIM) index, modulation transfer function frequencies (MTF 10 %) power spectra (NPS) values as parameters. Evaluation possible decrease image quality also performed by testing the homogenous water wire acquired mAs values. Results show that able consistently increase SNR, cPSNR, SSIM values, integral (NPS). However, level either training or data affects model’s final performance. lower tends result in over-smoothed images, indicated shift NPS curves. contrast, higher less satisfactory performance, Meanwhile, produce denoised reduced sharpness, MTF % Further studies are required better understand character RED-CNN for CT suppression regarding optimum parameters best results.