Q+A CENTER

Massachusetts General Hospital (Boston)
Mercury Computer Systems, Inc. (Chelmsford, MA)
(Nominated by Robert Murphy, Mercury Computer Systems, Inc.)

The Challenge:

Traditional mammography is based on two-dimensional X-ray imaging. One shortcoming of this modality is that tumors can be obscured and, therefore, go undetected. Digital breast tomosynthesis reconstructs a three-dimensional volumetric view from a series of eleven, low-dose, two-dimensional projection images. The technique enables physicians to “page through” the interior of the breast without obstruction by surrounding superimposed tissue. In the past, potential benefits of digital breast tomosynthesis were downplayed because the method was computationally intense and took far too long for use in a clinical setting.

The Solution:

Designers at Mercury Computer Systems implemented an iterative reconstruction algorithm called Maximum Likelihood Expectation Maximization which initially produced very high quality images. Engineers mapped the algorithm to a specialty graphics processor with a unique programmable rendering pipeline. Careful construction of the software port was required to overcome memory, bandwidth and instruction set limitations while optimizing run-time performance. The latter was reduced from five hours to five minutes. In pre-clinical testing, digital breast tomosynthesis permitted physicians to find cancers earlier and more easily, and to differentiate benign versus malignant lesions while simultaneously reducing false positives (which can occur with conventional mammography).

The Tools Used:

• Digital Breast Tomosynthesis imaging system, co-developed by Massachusetts General Hospital and Mercury Computer Systems
• GE Healthcare (Waukesha, WI) Senographe 2000 Digital Mammography System
• NVIDIA (Santa Clara, CA) Quadro graphics processor
• Maximum Likelihood Expectation Maximization algorithm, co-developed by Brandeis University (Waltham, MA) and Massachusetts General Hospital

The Difference It Made:

The ensemble of observed results, software implementation and acceleration hardware has confirmed that digital breast tomosynthesis improved cancer screening accuracy, lowered costs, increased patient comfort and reduced exposure to ionizing radiation.

In the United States, deaths from breast cancer dropped twenty percent over past decade, largely due to widespread screening mammography. Digital breast tomosynthesis’ superior ability to display lesions which otherwise might be masked could reinforce early detection and further reduce death rates. The Massachusetts General Hospital-Mercury Computer Systems technique sped up image reconstruction by a factor of sixty.