CHARACTERIZATION OF BREAST LESIONS BY PROCESSING DIGITAL BREAST IMAGES

This Rendering to the World Health Organization, women in both developed and developing nations are most likely to develop breast cancer. This illness causes breast cells to grow and multiply out of control. According to research institutes and international organizations, there are various screening methods available based on age, and breast cancer can be cured if detected in time. The Breast Imaging Reporting and Data System (BIRADS) is a standardized system that is commonly used in these techniques to report results and findings. Results are sorted by BIRADS into six categories, numbered 0 through 6. Furthermore, mammography is the most widely utilized screening technique. This study suggests using mammography data processing to identify breast lesions. Adaptive filters are used for image cropping and contrast enhancement during the pre-processing phase. The pectoral muscle is then segmented using segmentation techniques that consider morphological and area growth factors. The lesion is then divided into sections at the muscle and breast levels using the Discrete Wavelet Transform (DWT), which finds any micro calcifications. Furthermore, to distinguish between dense lesions and other kinds of lesions, an area cultivation approach combined with multiple thresholding techniques is employed. Lastly, the obtained segmentation is used to extract textural and morphological features. When expert-segmented and automatically segmented images were compared, the Sorensen Decade similarity index was 0.73, indicating the effectiveness of the suggested method. Considering that the lesion area on a mammogram can only be roughly delineated by hand or automatically, this is a promising outcome.据世界卫生组织（World Health Organization）统计，无论发达国家还是发展中国家，女性均为乳腺癌的高发人群。该疾病会引发乳腺细胞失控性增殖与生长。据多家科研机构及国际组织的研究结果，可根据年龄分层采用多种筛查手段，若能及时检出乳腺癌，该病症可实现治愈。乳腺影像报告与数据系统（Breast Imaging Reporting and Data System, BIRADS）是此类影像筛查中用于标准化报告检测结果与发现的通用体系，其将结果划分为0至6共六个类别。其中，乳腺钼靶摄影是目前应用最为广泛的筛查技术。 本研究提出基于乳腺钼靶影像数据处理的乳腺病变识别方法。预处理阶段采用自适应滤波器完成图像裁剪与对比度增强。随后，结合形态学与区域生长因子的分割技术对胸肌进行分割。再通过离散小波变换（Discrete Wavelet Transform, DWT）将图像按肌肉与乳腺层面分区，以检出微钙化灶。此外，为区分致密性病变与其他类型病变，本研究采用区域生长法结合多阈值分割技术。最后，基于所得分割结果提取纹理特征与形态学特征。 将专家手动分割图像与自动分割图像进行对比后，索伦森相似度指数（Sorensen similarity index）得分为0.73，证实了所提方法的有效性。考虑到乳腺钼靶影像上的病变区域仅能通过手动或自动方式大致勾勒，该结果已颇具应用前景。