Computer-aided Diagnosis For Breast Cancer With Fusion Of Human And Machine Intelligence

Breast cancer is the most common malignancy in females. The early detection anddiagnosis of breast cancer can reduces mortality and is a widely accepted practice in manycountries. Mammographic screening mammography, which have the advantage of convenient,high resolution, and repeatable, remains the most commonly used procedure for this purpose.Computer-aided diagnosis system as “the second readers” provides objective and effectiveauxiliary information for radiologists and plays an important role to reduce false negative rate,which becomes a research hotspot in recent years. However, The existing computer-aideddiagnosis (CAD) system is highly focused on the processes of the image processing, featureextraction and classification, rather than the experience and knowledge of the radiologist.Although the CAD system is a feasible way to diagnose breast cancer, in practice, the finalclinical decisions are made by radiologist due to the superiority of human vision indistinguishing benign and malignant masses compared with computer vision. The purpose ofthis study is to develop a fusion system that combines the attributes of human vision with thepowerful computational ability of computers for improving the accuracy of massclassification in mammography.In this paper, we study a new method to improve breast cancer diagnosis performancewith the fusion of human and machine intelligence. The follows list the main contributionsand innovations:(1) Developed a web-based platform for mammographic breast cancer diagnosis ashuman-computer interaction interface, which provided a tool for observer performance study.(2) Put forward a Positive Predictive Value (PPV) for each BI-RADS category torepresent the probability of malignant of each BI-RADS category diagnosed by differentradiologists when they observed the masses in mammography.(3) Two fusion model based on Alpha-Integration(AI) algorithm and Multi-Agent(MA)algorithm respectively were proposed to integrate human and machine intelligence for breastcancer diagnosis.(4) Developed a novel fusion system to integrate the decisions of CAD and radiologiststo obtain the final classification results. The experimental results indicated significantimprovements of our fusion system over the diagnosis of radiologist and CAD alone.