Medical imaging methods that emit no radiation are becoming more accurate for breast cancer detection. Two TU Delft researchers developed an algorithm that accurately reconstructs a three-dimensional image of a cancerous breast from ultrasound measurements.
Next week, the team, Dr. Koen van Dongen, an assistant professor, and Neslihan Ozmen-Eryilmaz, a doctoral student, both of the Laboratory of Acoustical Imaging and Sound Control in the Faculty of Applied Sciences, will present their new method at the International Congress on Acoustics in Montreal, Canada. The algorithm computes the reconstructed image as a function of the speed of sound. Their results show that with their method, the resulting 3D image is more true to the original than after using other reconstruction algorithms.
Three-dimensional images of scanned breast tissue are desirable because they help the doctor better distinguish the boundaries of structures than in a two-dimensional slice. The location of a tumor and the extent of its presence appear clearly in a 3D image.
Before van Dongen’s and Ozmen-Eryilmaz’s work, three-dimensional images from ultrasound scans were typically reconstructed with existing algorithms that had been developed for MRI and x-ray scans. However, these computational hand-me-downs produced blurry 3D ultrasound images because they did not take into account its specific wave phenomena, like scattering, refraction, diffraction, and interference. In response, contemporary researchers have produced ultrasound-specific reconstruction algorithms that are based on acoustic wave equations. Still, they were not perfect. Some were based on inexact estimations of the wave equation, and others were computationally heavy.
The TU Delft team calculated the speed of the reflected sound with a new set of equations by iteratively updating the speed profile until a cost function was sufficiently minimized. The resulting speed-of-sound profile then became the color-coded contrast values in the visible image. They tested their method on synthetic ultrasound measurement data that was based on a mathematical model of a full breast with a tumor. Compared to four other methods, van Dongen’s and Ozmen-Eryilmaz’s reconstructed image could best localize the tumor and most closely matched the contrast values of the original image. This is work has been includedin this year’s Journal of the Acoustical Society of America.
Ozmen-Eryilmaz, who defends her dissertation this year, said that she is really excited to have been invited to the conference. Later this year, she and her small family will adjust to a new situation, which might include more traveling. “When and if I go back to Turkey, it will be great to have a PhD from TU Delft,” she said.
Comments are closed.