Computer-Aided Detection Systems for Digital Mammography

Mammography is the typical diagnostic test for early detection of breast cancer. The presence of microcalcifications and masses in the images may be an indicator of the disease. The microcalcifications size is very small, so, in many cases, they are not visible from medical images by radiologists. On the other hand, masses can be also undetectable if image contrast is not good enough. The computer-aided detection (CAD) systems are useful tools in facilitating the physician´s diagnosis. The CAD system proposed in this work is aimed at improving image quality based on image processing. On these improved images, it segments the mammary gland and highlights the presence of microcalcifications and masses. The system improves image contrast by means of convolution filters, it also eliminates artifacts by means of morphological opening and closing and Laplacian filtering, and uses entropy-based methods for segmentation of the gland and morphological filtering and histogram readjustment to enhance microcalcifications in the image. Masses are detected using an iterative contrast increase method. The system was tested with an annotated database (DB) MIAS, in oblique lateral views of glandular, glandular-dense and predominantly adipose breasts, which included images of malignant and benign lesions and other breast images without them. The system was evaluated with respect to the DB annotation, for a sample of 115 images. The performance of the system revealed a sensitivity of 93.2%, a specificity of 85.3%, a precision of 90.4% and an accuracy of 92%.