Logo der Physikalisch-Technischen Bundesanstalt

Image quality assessment in mammography using neural networks


The correlation between predicted (log y) and true (y) values for different mammography devices are plotted. The closer the points to the diagonal the better is the prediction.

Artificial intelligence is a powerful tool to assist physicians in diagnosing diseases. We extended this line of research by applying modern machine learning methods to the field of image quality assessment for mammography. Mammography is an established diagnostic technique to detect early forms of breast cancer. However, to ensure that a sufficient image quality can be obtained with a minimal dose of radiation, the image quality of a mammography unit must be determined. 


We show with our method that neural networks are capable to assess the image quality of a mammogram. Our experiments cover various manufacturers and the proposed approach is shown to work across different neural network architectures. Both aspects highlight the generalizability of our method. Moreover, we train the neural networks solely with synthetically generated mammograms which makes their training feasible and data efficient.


This work was done in collaboration with the Dutch screening center LRCB.


This work was published in Machine Learning: Science and Technology (DOI Opens external link in new window10.1088/2632-2153/acf914)

Contact: Josua Faller (8.42)

Email: josua.faller(at)ptb.de