Computer Aided Detection for breast calcification clusters based on improved instance selection and an adaptive neuro-fuzzy network

In this paper, we propose an novel instance selection algorithm and an improved adaptive neuro-fuzzy algorithm for Computer Aided Detection (CAD) of mammography. Firstly, the X-Ray images are partitioned into blocks. Secondly, the texture model is built for all negative packages instances. The distances from the unknown instances to the average model of negative packages are calculated. The instance with the fastest distance is selected as the suspicious area. Afterwards, the main features of suspicious regions are extracted for classification. Specifically, we propose to use an adaptive neuro-fuzzy classification Linguistic hedge (ANFC-LH) algorithm for CAD. The experimental results show that this method not only has the ability to automatically extract Regions of Interest (ROI), but also can greatly reduce the computation time while keeping the detection performance. At the same time, the better accuracy rate and true positive rate are achieved compared with traditional methods.