CIS Student Enhances Digital Reconstruction of Tumors: Presents Findings at San Diego Medical Imaging Conference
By Amy Mednick
Biomedical researchers now have access to a more elegant method to digitally reconstruct microscopictissueslices, or histological sections, of tumor specimens into three-dimensional models thanks to the work of Shaohui Sun. Sun—a Center for Imaging Science graduate student—presented his findings in February at the SPIE Medical Imaging Conference on Image Processing in San Diego.
CIS Professor Nathan Cahill discovered the problem in a conversation with Nzola de Magalhaes, a RIT Biomedical Engineering professor who studies tumor vascularization. Magalhaes wanted to find a way to stack successive histological sections of tumors in chicken embryos to eliminate the usual distortion and registration problems associated with digital reconstruction of these images.
Cahill, who is also a faculty member in the School of Mathematical Sciences faculty, turned to his first-year graduate research assistant Sun to generate a mathematical algorithm to help solve the problem. “Shaohui spent a quarter learning about the limitations of prior techniques, developing the theory behind our new algorithm, implementing the new algorithm, and validating it on Nzola's data,” Cahill says.
Before he figured out the new algorithm, Sun says that Magalhaes used a much more laborious process of aligning the slices by hand. Previous techniques—extremely time consuming and difficult—produced only a course volumetric reconstruction of the tumors. Sun also needed to tackle the “aperture problem,” which is one of the main obstacles biomedical researchers face when attempting to digitally reconstruct three-dimensional specimens. When stacking 10s or 100s of these slices, the final result becomes twisted and distorted when examined next to the original specimen.
In the past, researchers have only looked at two successive images in the registration process. “I compared five to 10 images and then figured out the similarities between the slices. Once you have the matching features, you know what is in common and you can model a mathematical formula to solve the problem,” Sun says. Sun’s algorithm allowed him to compensate for rotational, scale, shear, and minute geometrical variations between the slices.
Cahill encouraged Sun to see this practical problem through to the end. “When Shaohui was able to establish that the new algorithm seemed to work well on Nzola's data (and performed better than more basic approaches), I knew that he had a good topic to submit to an international conference,” Cahill says.
Cahill and Sun wrote an abstract and submitted it to the SPIE Medical Imaging Conference over the 2011 summer. It was accepted for an oral presentation and, with Magalhaes, they wrote the full paper to submit to the conference proceedings. Sun, who is first author on the paper, presented his research at the RIT Graduate Research Symposium and at a seminar hosted by RIT's Center for Applied and Computational Mathematics. “By the time he gave the conference presentation, he had practiced enough so that he was able to do a great job,” Cahill says.
The work has useful applications, most directly, in research on tumor vascularization, but also in any application where histology is used such as cell growth and development, cancer research, and identification of pathologies, Cahill says.
Sun, 27, from Qingdao, China, says he appreciated the opportunity to attend the San Diego conference. “It was a great opportunity for me to gain experience from academic communities such as SPIE for my professional development, and it will also enhance RIT's reputation in the field of medical imaging.”
Sun is currently working with CIS Professor Carl Salvaggio on a remote sensing study involving virtual three-dimensional building reconstruction of the RIT campus and downtown Rochester.
Last Modified: 2:54pm 20 Mar 12