Algorithm Evaluation And Image Enhancement For Digital Breast Tomosynthesis Imaging

Breast cancer is the most widely distributed malignant tumor with the highest fatality among women around the world,which owns a morbidity of 7%～10%among all types of malignant tumors.The average onset age of breast cancer in China is earlier than that in western countries within 10-15 years.Among every 100 newly confirmed cases on the globe,twelve could from China.It is predicted that the number of new cases could be as high as 234,000 every year in China.Therefore,the breast cancer has gravely threatened the physical and psychological health of Chinese women and its early diagnosis and treatment are of very important clinical significance.X-ray digital breast tomosynthesis(DBT),could reconstruct volume image of breast tissue through multiple exposure in limited angle range.With the advantages of high-resolution,high contrast and high accuracy,DBT has the ability of clearly delineating the anatomies of fasciculi,cysts,and microcalcifications inside the breast.And therefore plays an important role in breast examination.Since the first commercial DBT machine recived the FDA certificate in 2011,DBT has been widely used in clinic.However,the advanced DBT imaging techniques are always manipulated by foreign manufacturers such as Hologic,Seimens and GE.To break the monopoly of foreign manuf2acturers,the supervisor of this thesis has coorperated with domain manufacturers on science and technology project to develop DBT machines and core technologies with full independent intellectual property rights.The work of this thesis is part of the project,which focuses on the DBT algorithm evaluation and image enchancement in the procedure of research and development.(1)DBT algorithm evaluationThis work employed three algorithms,including FDK,SART and ASDPOCS-TV,to reconstruct images from the projections of phantom as well as patient.The images were then comprehensively evaluated by comparing the signal-to-noise ratio(SNR),noise power spectrum(NPS),and artifact spreading function(ASF).The results reveal that,with the reduction of radiation dose,the noise and artifacts of the images are increased.The images reconstructed by ASDPOCS-TV and SART have their own advantages and disadvantages and are better than that reconstructed by FDK on noise and artifact suppression.Therefore,the exposure dose level and reconstruction algorithms need to be carefully selected for different clinical tasks.Contributions of this work include:we analyzed the performance and applicable scene of three algorithms,providing guidance for algorithm development in the DBT project;we used indices including the SNR,NPS and ASF to conduct comprehensive analysis,providing reference for the evaluation of other aspects of DBT system;we verified the effectiveness of low-dose simulation method,which can be applied in the simulation in the development of other algorithms.(2)DBT image enhancementThis work conducted personalized enhancement of DBT mammographic images based on deep learning(DL)technique.Specifically,we proposed the DL-ME(Deep learning mammography enhancement)model based multiscale architecture and residual network,and constructed style switchers for the input of DL-ME model to control the style of enhancement.The experimental results demonstrated that the multiscale architecture plays the key role in the enchancement of DL-ME.More,image with multiple enhancement styles can be easily acquired by a single DL-ME model with multiple style switchers.The contributions of this work include:with low-contrast,the accuracy of the raw DBT mammographic images for breast lesion inspection could not be high,which can be improved by enhancement;the variety of styles of images from DBT machines with different vendors causes difficulty to radiologists.To deal with this problem,we proposed an unified DL-ME model with personalized enhancement,providing radiologists with any style desired.Based on this work,the inflexible projection enhancement step in currently developed DBT prototype in the project can be upgraded to DL-ME personalized projection enhancement.