Research On Low Dose Breast Tomosynthesis Reconstruction Algorithm

Digital breast tomosynthesis is one of the important methods for breast disease screening.Based on X-ray Radiography,this technique uses a limited angle scanning method and combines conventional geometric tomography technique to reconstruct three-dimensional breast image,which can resolve tissue overlap problem and improve image quality.However,due to the existence of X-ray radiation,it may cause radiation damage to patients during the breast disease screening.Therefore,it is significative and valuable to reduce the radiation in DBT scanning.There are two main problems in low-dose DBT reconstructed images,one is the problem of noise in reconstructed images caused by reducing the X-ray radiation dose,the other is the problem of hardening artifacts around calcification caused by the existence of highattenuating substances.In order to improve the image quality of low-dose DBT reconstruction and speed up the reconstruction,we study from three aspects: reconstruction efficiency,hardening artifacts around breast calcification and reconstructed image noise.The specific research contents are as follows:(1)A GPU-based DBT reconstruction acceleration method is proposed to solve the low efficiency of DBT reconstruction.The algorithm uses CPU multi-thread programming technology and GPU programming technology to accelerate the DBT reconstruction algorithm.Compared with the CPU version of the reconstruction algorithm,the efficiency of the GPU-based DBT reconstruction algorithm is improved about 20 times.The success of this work has laid the foundation for other works in low-dose DBT reconstruction.(2)A low-dose image denoising algorithm based on dictionary learning is proposed to solve the problem of noise in low-dose DBT reconstructed images.Based on the traditional dictionary learning,the algorithm builds two dictionaries on the reconstructed image and the noise image respectively.Then the algorithm uses the reconstructed image dictionary and noise dictionary to decompose the low-dose reconstructed image.Finally,the algorithm obtains the reconstructed image without noise.Through testing on real data,the algorithm can effectively reduce the noise in the low-dose reconstructed image.(3)A hardening artifact reduction algorithm for breast calcification based on projection completion is proposed to solve the problem of hardening artifacts around breast calcification in DBT reconstructed image.Based on the traditional projection completion metal artifact reduction algorithm,a new segmentation method is proposed according to the characteristics of breast calcification,which uses the reconstructed images to locate the calcification and performs accurate segmentation on the projection images,which can efficiently and accurately segment breast calcification.Through testing on real clinical data,the effectiveness of the algorithm for hardening artifacts reduction is verified.