Microcalcification detection with morphological operations and watershed segmentation

Mammography is the main exam used for breast cancer early detection, allowing to find suspicious structures, such as microcalcifications, small calcium deposits that may be associated with breast cancer. Research in digital image processing in segmentation techniques aims to divide the image into regions based on specific characteristics to facilitate the visualization of structures and use them in classification algorithms. Given the challenge in microcalcification detection, which often appear as small bright spots and can be easily misinterpreted as noise, the goal of this work is to apply watershed segmentation in mammograms to segment microcalcifications, using morphological operators to detect internal and external markers and restrain the segmentation process. After applying different denoising methods, the marker detection step performs the h-maxima operation using a disk-shaped structuring element with a radius of 3 pixels and an iterative threshold to binarize the result. The outcomes of the marker-controlled watershed segmentation were labeled as true or false positives based on annotation provided by the image database. The comparison of the mean and standard deviation of evaluation metrics across different denoising techniques suggests that an appropriate combination between h-maxima operation, thresholding, and noise reduction method can be used for the detection and segmentation of suspicious objects with microcalcification characteristics.