Semiautomated breast cancer classification from ultrasound video

The present work presents a new approach, based on the elaboration of a whole video clip of the ecoghraphic acquisition describing the lesion from different viewpoints. The focus of the paper is in comparing performances which can be obtained, using a simple classification algorithm, using the full video clip against the more common single frame approach. The system consists of five modules: preprocessing, semiautomatic segmentation, extraction of morphological features, classification of each frame and integration to obtain the classification of the video clip. Results show that the data integration allows an increase in the correct classification rate from approximately 89%, on a frame-by-frame basis, up to 97% on the whole video clip.