USC Viterbi BME sophomore Arjun Karnwal developed an algorithm that saves significant time for radiation oncologists. Image/Arjun Karnwal
For biomedical engineers, it often takes many years of painstaking work before their research progresses from the lab and into clinical applications that benefit patients in the real world.
However, one USC Viterbi School of Engineering sophomore is already watching his research make a difference for young cancer patients in Los Angeles. Arjun Karnwal, an undergraduate student in the Alfred E. Mann Department of Biomedical Engineering, developed a software script that can significantly improve the way radiation treatments are conducted. It is currently being used by radiation oncologists at Children’s Hospital Los Angeles (CHLA), where it saves time and reduces errors for clinicians while maximizing the impact of treatment.
Radiation oncology is a therapeutic strategy for reducing, controlling or clearing tumors with high-energy radiation waves — a treatment used either on its own or in combination with other treatments such as chemotherapy. Radiation treatment can involve treating the entirety of a tumor with a uniform dose of radiation. Another promising treatment is spatially fractionated radiation therapy (SFRT), which attacks the tumor using ultra-high doses of radiation at preselected” hot spots” in the malignant tissue.
“I like to use a martial arts analogy for explaining SFRT,” Karnwal said. “So, think of when you’re trying to take down a person. You can flick them everywhere on their body in a uniform approach. SFRT is like choosing a couple of weak points on a person and putting all your power into those instead of just uniformly attacking everywhere.”
SFRT also holds significant advantages for patients. Targeting a high dose of radiation to select points within a cancer’s mass can result in less damage to the surrounding healthy tissue. However, the challenge for clinicians is maximizing the number of hotspot targets in the tumor mass while keeping the radiation waves at a safe distance from healthy tissue.
“We don’t want these points to be too close to each other because the radiation can compound, and we don’t want that to happen,” Karnwal said. “In addition, we don’t want them to be too close to the edge of the tumor just for extra peace of mind.”
Karnwal said that another way to imagine the problem was to compare it to social distancing guidelines from the early days of COVID-19.
Left: Karnwal’s algorithm-generated vertices for radiation hotspots in a tumor mass. Right: a comparison of algorithm-generated vertices (yellow) and human-generated vertices (red). Image/Arjun Karnwal.
“You can think of the challenge as if you’re trying to maximize the number of people in your house given COVID guidelines (six feet apart in all directions). This is easy if your house is a perfect box. However you also have tables and chairs that make maximizing the arbitrary space difficult,” Karnwal said. “It also gets complicated when we’re worried about three dimensions. You can be too close to someone, say, on the floor above you or below you, too, that’s an extra factor to take into account.”
However, planning a safe maximum number of radiation hotspots for tumor treatments can be quite a time-consuming process — one that currently needs to be manually mapped by the radiation oncologist. The challenge is that tumors vary wildly in shape, and one oncologist’s proposed map of where to focus radiation may differ significantly from another’s map.
When Karnwal was observing the workflow in the radiation oncology department at CHLA, his colleagues posed this problem: is there a more effective way to achieve an optimal map of radiation hotspots in individual tumors? Karnwal could see that the process would benefit from a more iterative, randomized system to optimize the arbitrary volume of hotspots.
“There’s other software out there that currently has alternate methodologies. One of them makes a grid within the tumor, and that would work great if your tumor was a perfect rectangular box. But typically, tumors have atypical, unpredictable geometries, and I just thought the grid style wouldn’t yield the maximum number of targets,” Karnwal said.
“That’s what got me thinking about randomness — the more sophisticated engineering word for that is Monte Carlo simulation — which is basically running something a bunch of times, and there’s some degree of randomness in there. Then we use that to simulate potential outcomes.”
Karnwal’s solution was to write a software script that works with the current SFRT mapping program and allows clinicians to input their preferred parameters before calculating a range of randomized approaches that would be most effective for individual treatment. The system would then select the map that yielded the maximum number of targets within the tumor.
Karnwal’s script has saved valuable time for CHLA clinicians. Where it previously took an average of two hours to build a comprehensive radiation hotspot map manually, now they can do so in just eight minutes.
“This also enables same-day treatment, which was previously impossible because it takes so long to plan and approve treatment,” Karnwal said. “Time is the one thing we can never get back, and it means the world to me to have saved so much of it for so many people.”
After about nine months of trial and error in the algorithm’s development phase, the feedback from Karnwal’s clinician colleagues has been overwhelmingly positive. Now, they have a tool that allows them to plan out comprehensive radiation treatments quickly and with confidence.
“One thing I’m super proud of – it’s an algorithm, so when we give it some constraints, it’s always going to follow those constraints. That means that there are zero geometric placement violations. The targets will always be far enough away from healthy tissue and from each other. There’s no doubt that these spots are valid, and therefore, we can promote the best patient outcomes,” Karnwal said.
The results have proven so effective for the CHLA team that Karnwal’s script is now also being used at USC Norris Comprehensive Cancer Center to save time for radiation oncologists. He hopes to continue consulting with clinicians and other stakeholders to further refine his script so that it is faster, more intuitive and easier to use. Karnwal recently presented his work at the 2025 Radiosurgery Society Annual Meeting in Arizona, spotlighting the latest developments in radiotherapy techniques.
“I’m so grateful for this opportunity. Engineering projects typically take a long time and require a lot of perseverance,” Karnwal said. “This project took many weeks of just me saying, ‘Hey, does this work?’ and a lot of back and forth. It just makes all of that iteration worth it in the end. Knowing that you can contribute in a meaningful way is super powerful and inspiring.”
Published on March 27th, 2025
Last updated on March 27th, 2025
