Machine learning for noise removal on breast ultrasound images

This study assessed the utility of machine learning for isolating noise and artifacts in breast ultrasound images. Such corrupt image regions (ROIs) can be automatically excluded when registering images acquired from different angles. Artifacts included posterior acoustic shadowing and enhancement arising from cancers and cysts respectively. Images were obtained on a breast-mimicking phantom containing multiple cysts and lesions with variable speed of sound and attenuation properties. In vivo breast images of cysts and cancers were also available. Results show that the classifiers were able to identify the regions of corrupt data accurately.