Abstract: ROI(region of interest) detection for mammography is the first step of CAD(computer-aided diagnosis) for breast disease. Mass is the region what we are interested in mammography, so the purpose of this paper is to improve mass detection because that will help reduce the rate of misdiagnosis. Lots of work has focused on this field these years. Traditional methods only extract single visual feature for classification to detect the mass. However, mammography is complex. Employing only single visual feature will ignore the connection between images and features. That’s why context information of mammography should be taken into consideration. Besides, feature integration is another important method to make up such deficiency which has been implemented in different ways and achieved good results recent years. But few work has considered both context information of mammography and feature integration method to improve ROI detection. As a result, in this paper a ROI detection method named MCID(multi-cue integration detection) is proposed based on multi-cue feature integration. The core idea of this method is integrating correlation feature of context and several visual features together. First of all, both global and local correlation features are extracted by using the proposed multi-cue PLSA model (MCPLSA), which is an improvement of PLSA, and then a classification model named CFIBM(correlation feature integration based boosting model) is proposed in order to improve the accuracy of mammography ROI detection. This model is made up of several logistic regression classifiers whose input has taken context information into consideration and output is the probability of being predicted as each label(mass or not). At last, the final result is voted by each classifier. Experimental results on DDSM(digital database for screening mammography) show that MCID performs better than existing methods in two aspects, one is improving the accuracy of mass detection and the other is shortening the time needed.