Research On Medical Image Classification Based On Bag-of-Words Model And Association Rules

Recent years, as the continuous penetration and development of computer technology and image processing technology, computer-aided diagnosis shows its increasing importance in doctors' diagnose and treatment of diseases, and has become one of the significant medical research directions in the world. In this paper, deep analysis, research and experiment are carried out for the efficient classification of mammogram images.Related research works in this paper are as follows:Firstly, based on traditional BOVW model, an improved max frequent item-sets based on BOVW(MFI-BOVW) method is proposed. The method gives the visual word with space and semantic information, and improves the similarity degree of the same kind of the visual word histogram, and makes the difference between different categories more significant. This improved method was used in the experiments on MIAS mammogram image set to achieve the preliminary classification of the lesions and normal images and get suspicious lesions.Secondly, based on the traditional association rule classification algorithm CMAR, considering the influence of attribution and lift degree, an improved algorithm SR-CMAR is proposed which is suitable for the medical image classification. After preliminary classification of suspicious lesion, improved algorithm is used to find suspicious image area. The experimental results show that the improved algorithm can effectively improve the classification efficiency and ensure the classification accuracy.Based on the achievements of domestic and foreign scholars, this paper uses bag-of-visual words model and association rules for mammogram image classification after preprocessing and feature extraction. Experimental results also demonstrate the effectiveness and feasibility of mammogram image classification based on bag-of-words model and association rules. In addition, general lesion areas could be located roughly at the same time.