Breast X-ray Images, Computer-aided Diagnosis Algorithm Research

Mammography is the first choice to diagnose mammary diseases, especially the breast cancer. It is of great importance in medical diagnosis and research to study mammogram. The quality of mammogram is poor for the nature of breast tissue. A computer-aided image analysis technique can provide a consistent and reproducible econd opinion?to a radiologist, which may reduce falseegative diagnosis Clustered micro-calcifications and masses are the two most important indications of malignancy on mammogram. Some newly algorithms have been adopted to detect the micro-calcifications and masses. A BP neural network has been employed to micro-calcification detection. Firstly, a new multi-tolerance region growing method is proposed for the detection of potential calcification and extraction of their contours. Then we pick up the compactness, moments, Fourier Descriptors and contrast characters of the segmented region. Finally, a three-layer BP neutral network is utilized for the classification of calcifications. In our study, the true positive (TP) rates of test set (96 samples) were approximately 89.5%, and the false positive (FP) rates is 7.3%, false negative rates is 4.2%. A region growing method based on fuzzy set has been introduced in mass detection. And directional factor-traversing algorithm has been used to pick up the outer contours of the mass. Then we select four region features: average, standard deviation, contrast and compactness. Finally, a T-detection has been applied to verify the region growing result and satisfied results are reached. At last, we combine the tumor and calcification detection with a mammogram database manage system. A completed computer-aided mamniogram management and diagnosis system has been built and applied successfully in mammogram research.