icones-sociais-facebook icones-sociais-linkedin

logo-ibeb-side-area

Improvement of image quality and dose reduction in digital breast tomosynthesis using statistical image reconstruction algorithms

Tomosynthesis is an emerging radiological technique used for breast imaging. It produces a 3D image from a series of views, reducing the problem of overlapping structures in 2D mammography imaging. The clinical benefit of this technique for the diagnosis of breast cancer is still under evaluation. Recent studies show that Digital Breast Tomosynthesis (DBT) presents superior image quality and better cancer visibility, indicating a better sensitivity than digital mammography (DM). Some argue that DBT has not yet proven to result in better clinical performance than using DM, but others suggest that combining both techniques could lead to better overall performance. However, some authors go one step further and consider that DBT can be an alternative to DM in breast cancer screening, particularly for women with dense breasts. Despite these promising results, it is acknowledged that neither the acquisition nor the image reconstruction algorithms of DBT are fully optimized.

The majority of published clinical studies refers the use of analytical algorithms for image reconstruction. However, it is accepted that iterative methods can improve image quality relatively to analytical image reconstruction.

Iterative methods include statistical iterative reconstructions algorithms that incorporate a geometrical model for the transmission/detection process and take into account data statistics and a noise model. The inclusion of this statistical information enables better quantitation than Filtered Backprojection (FBP) in regions of low intensity / low count rates, as demonstrated for emission tomography.

In the proposed work, this characteristic is crucial since the number of photon counts in DBT projections is reduced as a consequence of low dose requirement. Therefore, we anticipate that statistical algorithms can not only lead to better DBT images, but can also play an important role on dose reduction if similar image quality can be obtained at a lower dose, increasing the clinical value of this technique.

Hospital da Luz has installed the first DBT scanner in Portugal and, being a research partner of this project, grants us access to the equipment. This is crucial not only for simulation validation and dosimetry purposes, but also for the development of image reconstruction algorithms, since the access to phantom and real data is mandatory along the way. In addition, radiologists from Hospital da Luz will contribute to the evaluation and optimisation of the new or redesigned algorithms.

In order to evaluate the potential reduction of patient doses, we will develop a Monte Carlo simulation platform, similar to the one that has already been validated for conventional digital mammography. This simulation tool will allow us to test different exposure configurations, for both dose and image quality purposes, without the need to irradiate patients. This simulation platform will be validated with dose measurements made with conventional dosimeters and with a new dosimeter that is currently being developed by LIP, one of the research partners.

Regarding image reconstruction, an important issue that hinders the use iterative image reconstruction algorithms is their computation time, which can amount up to several hours in a standard computer. To overcome this problem, we intend to take advantage of the modern graphic processor units (GPU) that are known to speed up image reconstruction by more than an order of magnitude. Some work has already been done within the IBEB group.

In recent years, this team has published several papers in image reconstruction and processing, Monte Carlo simulation and dosimetry. The two radiologists from Hospital da Luz have a long experience in mammography and are using DBT in clinical routine for more than two years. This background encourages us to address the problem from an integrated perspective. We do believe that there is work to be done in the improvement of image reconstruction algorithms for DBT and that this can have a significant repercussion on dose received by the patients. We also believe that simulation can provide additional information to radiologists and radiographers, so they can optimise on everyday routine each particular exam in respect to dose deliver to patients.

This project congregates academic and clinical expertise. The expected results are due to be fully available to the clinical partner, for the benefit of its patients. The global impact that DBT is expected to achieve in breast cancer diagnosis increases the potential value of the tools developed in this project. In the case of commercial valorisation of these tools, the institutions have already signed an agreement that includes the terms of the sharing of the intellectual property.

Principal Investigator: Pedro Almeida (IBEB)