Data-driven models for inverse problems in GI-BCT
Breast cancer is the most common malignancy in women. Unfortunately, the breast imaging technologies that are routinely used in clinical practice are characterized by important limitations, such as poor soft tissue contrast, low resolution or limited to 2D imaging. To fill the gap in todays breast imaging, a grating interferometry breast computed tomography (GI-BCT) prototype has been designed and is currently built by our group.
At synchrotron beamlines it has been shown that phase contrast CT, and in particular also grating interferometry CT, is able to produce high quality tomograms of biological soft tissues. However, the need for a sufficiently large field-of-view to accommodate an entire breast, and thus large gratings which are difficult to fabricate, and the impossibility to use a synchrotron source in a hospital (less coherence and finite bandwidth) make it significantly more challenging to obtain high quality data. Moreover, clinically imposed constraints regarding radiation dose and scanning time demand for sparse sampling and set an upper threshold on the photon count statistics, thereby making it even more cumbersome to achieve high image quality.
This project focuses on the development of data processing and reconstruction pipelines to generate high quality reconstructions from sparse and highly noisy data obtained with the scanning protocols foreseen for first clinical tests. In particular, given the challenging imaging conditions, this project attempts to achieve this goal by developing customized data-driven deep learning algorithms to tackle the challenging ill-posed inverse problems that arise in grating interferometry breast computed tomography.