Recognizing breast tumors based on mammograms combined with pre-trained neural networks

Breast cancer is one of the most common cancers in women worldwide, and it seriously threatens people’s lives and health. Breast Imaging Reporting and Data System is developed as a standardized system or tool for reporting breast mammograms, where different grades of diagnosis and treatment are critical to the survival rate and survival time of patients. Efficient computer-aided diagnosis of breast tumors based on computer vision models can better assist physicians in selecting effective treatment options, thereby reducing patient mortality. Therefore, early detection and early treatment are of great significance to patients with breast disease. In this study, a new image enhancement framework, called Image Negatives and Contrast Limited Adaptive Histogram Equalization Image Enhancement, was created for the first time based on the comparison of a set of multiple data preprocessing methods for detecting normal, benign, and probably benign breasts. The ResNet-50 pre-trained neural network was used for feature extraction and the classification results were compared on K-nearest neighbor, Random Forest, and Support Vector Machine classifiers. The evaluation indexes adopted in this paper include confusion matrix, precision, sensitivity, F1 Score, etc. These evaluation indexes can be used to evaluate the model in a very comprehensive and accurate way. The experiments show that the KNN classifier has the best classification result, the classification accuracy is 85%, and the AUC is 0.89. It is proved that mammography, as a non-invasive screening tool, has certain practical significance in effectively evaluating tumor grade and its clinical application.