Computer-aided Mass Detection Based On Ipsilateral Multi-view Mammograms

Being the second most fatal disease of woman,breast cancer has aroused widespread attention in human society.To improve the objectivity and accuracy of the diagnosis results and assist doctors to make the final diagnosis,image processing and machine learning methods are used to detect abnormities automatically in the computer aided detection(CAD)system.As one of the main symptoms of breast cancer,mass detection is of core issues in CAD system.To date,most of the development of CAD system has been based on the analysis of single mammogram in clinics.However,the results of current CAD system based on single-view mammogram have many false positive regions,and the radiologists could not have confidence in accepting this type of schemes.In this paper,CAD system based on multi-view mammograms is proposed to improve the performance of single-view methods.Considering that the multi-view-based method is produced on the basis of the single-view ones which may affect the result,single-view-based method is studied at first.Then the multi-view-based method is completed.In summary,specific contents of this paper can be summarized as follows:Firstly,this paper integrates visual saliency model and deep learning techniques to form a CAD system which is based on single-view mammogram.In detail,frequency-tuned saliency detection worked well in reserving the details of mass,and it improves the contrast between masses and normal tissues.To avoid handcrafted features and reflect the characteristics of mass and normal tissues,the deep learning framework Caffe and finetuned Alex Net model is used to deep representations of mammography.The results of single-view-based method remain a high detection rate,which provides a good basis for the realization of the CAD system based on multi-view mammograms.Secondly,since projected distance to the nipple along the centerline is invariant,matched regions carried out on both ipsilateral views(cranio-caudal view and mediolateral oblique view)are established.Then,the deep features are extracted and connected for each subimage of a matched pair,and they are finally propagated to the Support Vector Machine(SVM)classifier.The performances of CAD system based on both single-view mammogram and multi-view mammograms are evaluated and compared.The experimental results show that multi-view-based method could maintain the mass detection accuracy while reducing false positive rate of the single-view-based method.The multi-view-based method could improve the performance of the CAD system based on single-view mammogram.The researches on multi-view-based method and single-view-based method in this paper establish theoretical basis for wide application of CAD systems in clinics.