Tool for Semi-Automatic Segmentation and Labeling of Regions of Interest on Mid-Lateral Oblique Screen-Film Mammograms

Artificial intelligence (AI) algorithms have enormous potential applications and cover the entire medical imaging lifecycle, from image creation, diagnosis, classification and even prediction of particular pathologies. A problem for the development and clinical application of AI algorithms is the limited availability of extensive and representative training data including expert labeling (Manual Labeling). Current supervised AI methods require a data pre-processing stage for labeling data to train, validate and test algorithms with an optimal approximation to the ground truth. In this paper, a graphical interface was developed and implemented to label in a semi-automatic way Regions of Interest of Mid-Middle Lateral Oblique Screen-Film Mammography (MMLOSFM) through pre-processing techniques such as segmentation and image enhancement. Based on the analysis performed using four structural comparison indexes, it was demonstrated an efficient segmentation and labeling of MMLOSFM with a consistently high similarity (91.6% compared to the ground truth provided by the MIAS database).