Image Reconstruction with Regularization for Small Number Projection Tomosynthesis

Abstract

Tomosynthesis is the technique of reconstructing 3D images from a projection within limited angles. The device under development with four fixed X-ray tubes makes it possible to perform imaging for about 1 second and this also makes it easier for the patients to hold their breath due to the briefness of the imaging process. However, the image quality of the reconstructed image tends to be degraded because the number of projections is small. In this study, the availability and effectiveness of using an image reconstruction algorithm with regularization for small projection tomosynthesis was examined. The imaging of a digital phantom and a measurement chest phantom, reconstructed by ML-EM and ML-EM+TV, was compared using RMSE and CNR. In the numerical simulation, the image reconstructed by ML-EM+TV makes it possible to suppress the artifacts and thereby improve the resolution in the depth direction. According to the measurement data, TV regularization makes it possible to suppress the noise and improve the CNR by increasing the number of iterations. Therefore, image reconstruction with regularization for small projection tomosynthesis was found to effectively reduce both artifacts and noise.