After rapidly typing code for the third straight hour at his cubicle in the Laboratory of Imaging Informatics of Dr. Daniel Rubin, senior Rishi Veerapaneni took a moment to glance outside at students playing frisbee on the manicured lawns of Stanford University’s campus on a hot August day.

Soon, his summer research internship would be nearing an end, and in September he would be submitting his research project to the team category of the annual Siemens Science competition. Little did he know that in his first year applying to Siemens, his project would go on to become one of 466 semi-finalists from a pool of nearly 1,800 submitted projects.

While the glory and prestige of the Siemens science fair attract thousands of students every year, Veerapaneni’s original intent was not to enter Siemens but to simply conduct cutting-edge research and gain valuable lab experience.

“If you do research over the summer, you hope to at least submit a paper,” Veerapaneni said. “If the Siemens competition makes sense, you go from there.”

In mid-March of his junior year, Veerapaneni applied to different summer science programs and reached out to professors at local universities for research opportunities. He sent 125 emails and received 47 responses, some of which had compelling offers for summer internships.

He decided to work at the Laboratory of Imaging Informatics at Stanford for Rubin. During the internship, Veerapaneni conducted his research with partner Arjun Subramaniam, a junior at Harker. The pair was mentored by Assaf Hoogi, a postdoctoral student at Rubin’s lab.

Veerapaneni’s project focused on coding a complex computer algorithm that would automatically locate and outline the boundaries of cancerous tumors using computer vision and machine learning. Subramanian handled the machine learning aspects of the project and Veerapaneni used the machine learning results for the actual computer vision algorithm.

“Outlining cancerous lesions in medical images is an important step toward the diagnosis and treatment of cancers,” Veerapaneni said. “Yet doing so is a difficult task because [these lesions] come in a variety of shapes, sizes, and contrasts.”

To overcome this difficulty, Veerapaneni and his partner used level-set, a method widely used for lesion outlining, or segmentation. Level-set begins with an initial circular outline, which is then modified to contour the edges of the actual lesion. Veerapaneni and Subramanian worked on improving the level-set process, which is hindered by problems such as noisy images, in order to control the expansion and contraction of the outline.

“We trained a convolutional neural network (CNN) to predict whether the location of the contour was inside, near the boundary or outside the lesion,” Veerapaneni said. “With these predictions, we could adaptively adjust the level-set parameters, which in previous projects were kept fixed and considered unimportant.”

The team’s new method, called AdaptSet, proved far more accurate and consistent than level-set. Veerapaneni tested AdaptSet on an MRI image dataset of liver lesions, and found that his method was better able to deal with inaccurate initial contours, noisy images and low-contrast lesions.

Throughout the summer, Veerapaneni and Subramaniam spent five days per week for about 11 weeks at Rubin’s lab from about 10 a.m. to 5 p.m. with an hourlong break for lunch, perfecting their algorithm.

At the beginning of the internship, Veerapaneni read instructional papers on the lab’s research that established the expected research standard and gave him a general picture of the problem he was trying to solve. From there, he and his partner embarked on its long research process.

A typical day included trying out different ideas on the algorithm, a painstaking process of trial and error.

“My daily routine was writing the code, testing the code and thinking of a new idea or a new improvement to try out,” Veerapaneni said. “The process was at times frustrating, but ultimately very rewarding.”

For one hour on Thursdays, the team had meetings in which postdoctoral students presented the different projects they were working on. To balance the demanding research work with some leisure time, Veerapaneni joined Stanford’s juggling group, which met every Friday over the summer at 5 p.m.

“I like juggling because when I’m juggling I can’t really think of anything else,” Veerapaneni said. “It’s a way for me to ignore the worries and stresses of life. Partner juggling with the Stanford students and trading stories with them was really fun.”

Throughout this school year, Veerapaneni has shown his project to math and computer science teacher Debra Troxell on multiple occasions to keep her updated on his progress. Troxell is impressed with Veerapaneni’s research and its implications in the medical field.

“[He] has always been a highly motivated young man with the initiative to take his educational interests to the next level and change the world,” Troxell said. “How amazing is a project that can define the edges of tumors in a new way? It can fundamentally change the way doctors diagnose their patients.”

At the end of the internship in August, Veerapaneni and Subramaniam decided to apply to Siemens. They finished the majority of their research work prior to the start of the school year. After the school year began, Veerapaneni and his partner dedicated most of their time to co-writing the research paper required for Siemens.

According to Veerapaneni, the Siemens application is straightforward and does not require personal essays or teacher recommendation letters, as other prestigious competitions such as Intel Science Talent Search (STS) do. The only part of the Siemens application that is judged is the research paper.

“At first, writing 18 pages seemed kind of daunting,” Veerapaneni said. “But by the end we actually had to shorten our paper because we had gone past the page limit.”

Although Veerapaneni did not advance to the Siemens competition finalist round, he felt that his summer internship was an invaluable experience, and that his research was able to help Rubin, the university and the entire scientific community. In college, Veerapaneni hopes to pursue a computer science or electrical engineering major where he may apply his computer skills in meaningful ways.

“The internship opened my eyes to the different applications of CS, and showed me a different side of developing new technologies,” Veerapaneni said. “Nowadays, people usually correlate new technology with snazzy tech projects like Google’s self-driving car. But research in academia is also an exciting realm that I hope to continue to be a part of.”