Quantitative evaluation of total variation noise reduction algorithm in CT images using 3D-printed customized phantom for femur diagnosis

The purpose of this study was to evaluate the usefulness of the total variation (TV) noise reduction algorithm for improving image characteristics with 3D printing technique for the accuracy enhancement of femur fracture diagnosis with computed tomography (CT) images. For this purpose, 3D printing technique was applied to output customized phantom, which can copy a human femur, with iron-polylactic acid filaments that has similar density of human bones. Then, CT images of 3D printed customized femur phantom were obtained by applying the conditions for tube current of 100 mA, 200 mA, and 400 mA, respectively. In addition, a TV noise reduction algorithm applied with regularization parameters of 0.005 and 0.05, respectively, was modeled and applied to 100 mA CT images. To quantitative evaluate the degree of improvement for TV noise reduction algorithm, coefficient of variation (CV) and contrast to noise ratio (CNR), which are noise level evaluation parameters, and blind/referenceless image spatial quality evaluator (BRISQUE) and natural image quality evaluator (NIQE), which are no-reference image quality assessment, were calculated. As a result, when the TV noise reduction algorithms with regularization parameters of 0.005 and 0.05 were applied to the 100 mA image, the noise level evaluation parameters showed similar or improved to the 200 mA and 400 mA images, respectively. On the other hand, the BRISQUE was most improved when the TV noise reduction algorithm with regularization parameters of 0.005 was applied, however, the 0.05 showed poorer results than the images without algorithm. In addition, the NIQE showed that when the TV noise reduction algorithm was applied, the results were improved compared to the CT images without the algorithm, and the regularization parameter of 0.005 was improved than 0.05. In conclusion, we demonstrated through experiment with 3D printing technique that TV noise reduction algorithm with an appropriate regularization parameter can solve the problem of noise and artifacts in the femur CT images.