Research On Mass Recognition Algorithms In Mammograms

Breast cancer is one of the common malignant tumors and remains the leading cause of cancer death among females. Clinical experiments demonstrate that early detection and early treatment can increase the potential of survival. Mammography is a preferred method and also the most efficient and reliable tool for early prevention and diagnosis of breast cancer. Breast mass is an important symptom in mammograms. However mammograms are always with low contrast and the mass boundaries are blurry, which make it difficult to discover all the disease in time. With the development of computer technology and the support of various advanced theory, computer-aid diagnosis (CAD) has become the international research hot spot worldwide. The CAD system can efficiently decrease both the rate of missing and erroneous test, which offers the doctors a reliable second suggestion.In this paper, we mainly studied on breast mass segmentation and recognition method, the main contents are as follow:1. We propose a mass extraction method based on the Vector Field Convolution (VFC) Snake model and the Rough Set theory. According to the characteristics in mammograms, firstly we use the Rough Set theory to enhance the contrast and obtain the mass location by Hough transformation algorithm. Next, the contour can’t deform to the mass boundary completely when we segment the mass applying the VFC Snake directly, thus the algorithm is improved from two aspects:the edge map is calculated by performing the Canny operator on the Rough Set method amended image and then convolute with the vector kernel as a new external force, besides employing the mass location information as the model internal force. Experiments show that the improved method achieves higher detection rate, accuracy and more similar to the manual-segmented results.2. The mass recognition algorithm is presented based on the mass region features and the Random Forest classifier. Following the mass segmentation procedure, we extract the shape, margin, intensity and textures features from the mass region and its surrounding and establish features database with 32 dimensions, and then input these features to the Random Forest to distinguish the mass into benign or malignant. At the same time, the Support Vector Machine (SVM), Genetic Algorithm Support Vector Machine (GA-SVM), Particle Swarm Optimization Support Vector Machine (PSO-SVM), Decision Tree classifiers are used as comparisons. The results show that our method achieves the best performance with accuracy of 93.24% and 91.73% in unique DDSM database and the both two database respectively, and the area under the Receiver Operating Characteristic (ROC) curve reaches 0.9632 and 0.9467 respectively.Experimental results prove that the proposed method could provide some theoretical basis to the design of computer-aided mass detection and diagnosis.