Research

Many applications in breast imaging such as dosimetry models, x-ray scatter correction algorithms, and image processing techniques require realistic virtual breast models. We aim to provide objective models for use with these methods. As a first step, we have created a realistic 3D model of the shapes of compressed breasts undergoing mammography or breast tomosynthesis. For this, a protocol was developed to accurately capture the 3D shape of patients’ breast before and during compression for mammography. Using a pair of 3D surface scanners, the patient breasts are scanned in three conditions, resulting in accurate representations of the breast surfaces before, during, and after mechanical compression by the mammography system. From these data, we aim at developing a computer model that can generate a representative sample of 3D compressed breast shapes, with potential application in the improvement of digital breast tomosynthesis image reconstruction, and in the validation of computer algorithms that simulate mechanical deformations of the breast. The end goal for this latter application is to create and validate an image processing algorithm able to track the tumor position from x-ray images for different breast positions (e.g. prone and supine position), with the improvement in surgical planning as a potential outcome.

Camera setup of the structured light scanning system, with a phantom placed on the breast tomosynthesis system.

Post processing steps of the 3D external breast surface, from left to right: Raw surface images acquired by the two surface cameras; cleaned images with signals that do not belong to the breast surface removed; and merged surfaces from both lateral cameras after fusion using global fine registration.

These two images show the difference between a standard tomosynthesis reconstruction (left) and a reconstruction which includes the information from the surface scan of the compressed breast (right). Adding this information recovers the attenuation values near the breast edge and allows for better visualization of that region.

Researchers:

Marta Pinto

Koen Michielsen

Ioannis Sechopoulos

Key Publications:

•  K. Michielsen, A. Rodriguez-Ruiz and I. Sechopoulos. "Estimating the compressed breast-shape using deep learning.", 2020. Abstract. DOI.
• A. Rodríguez-Ruiz, G. Agasthya and I. Sechopoulos. "The compressed breast during mammography and breast tomosynthesis: in vivo shape characterization and modeling.", 2017. Abstract. DOI.