Wednesday, 26. May 2010 26.05.10 12:00 Age: 8 Jahre

Replacing the second radiologist
PC² accelerates automatic breast cancer diagnosis

By: Katharina Bätz, English translation: Katrijn van Oudheusden

Tobias Beisel in front of an edited MRI-scan. (Photo: Katharina Bätz)

Researchers of the Paderborn Center for Parallel Computing (PC²) at the University of Paderborn have set themselves a challenging goal: the team around project leader Prof. Dr. Marco Platzner and Tobias Beisel, research assistant, is cooperating with the software company CADMEI GmbH in Bingen, Gernany, to develop an automatic diagnostic system for magnetic resonance imaging (MRI). The system is intended to provide support for radiologists in examining the female breast for tumors, and in the longer run the system could even provide an instant second opinion.

MRI is an imaging technology used to visualize the detailed internal structure of the body’s organs and tissues. MRI uses electromagnetic pulses in a powerful magnetic field to agitate the hydrogen atoms in the body’s cells. The amount of available hydrogen atoms and the fading pattern after agitation, allow radiologists to recognize and visualize tumors in the body.

MRI is a very time-intensive method: both the time needed for imaging in the MRI machine as well as the image generation and editing phase afterwards take a lot of time. In the current practice of manual image editing, a radiologist assesses the images produced by the MRI machine using a simple image editing program, and then uses this information to make a diagnosis. In automatic image editing, the lack of focus in the MRI data is calculated out of the images automatically using complex algorithms. Tissue areas that look suspicious are identified automatically using specific parameters and are classified automatically into malignant or non-malignant tissue. All of these measures can help the radiologist make a more specific and confident diagnosis.

"We cannot accelerate the MRI procedure itself, but we can accelerate the editing phase using parallel graphics computing", explains Tobias Beisel. Graphics processing units were developed for use in computer games, but due to their massive parallel processing and their good price-performance ratio they are used more and more in high performance computing. Based on their parallel computing experience, the researchers in the PC² group have been working on improving the speed of MRI-data editing and diagnosis-automation since July of 2009.

Currently, a radiologist requires around half an hour per patient for the evaluation of the MRI-data, and then informs the patient about the result. A second opinion of another radiologist is not absolutely necessary, but is best practice in renowned international clinics. If a second opinion leads to a different diagnosis, the patient needs to undergo a second appointment and a second round of stress to resolve the situation. The automatic diagnosis using CADMEI GmbH’s software still takes between one and multiple hours. The goal of the research group is to bring this down to around 10 minutes. In this way, the computer-aided diagnostic system could support radiologists during their very first interpretation of the MRI-data. A second step would then be the automatic generation of a second opinion by the diagnostic system after a medical valorization. Tobias Beisel notes the advantages of the faster computing: “The patient could then be examined in the MRI and immediately informed of the double-checked diagnosis afterwards by the doctor. In this way the patient is spared long stressful waiting time for the final results.”

The project is funded by the German Federal Ministry of Economics and Technology as part of the central innovation program SME (small and medium-sized enterprises). A ready-to-market computer system that allows for computer-aided tumor diagnosis will be developed by mid-2011.

Contact:
Tobias Beisel
Paderborn Center for Parallel Computing (PC²)
+49 5251 60-6327
tbeisel[at]upb.de