Qu-Net at USC: Quantum-Enhanced AI Draws Sharper Tumor Boundaries With Fewer Parameters

Quantum-enhanced imaging for sharper tumor boundaries

While many researchers race to prove what quantum computers can do faster than today's machines, Amir Kalev at USC Viterbi's Information Sciences Institute is focused on a different question: How can quantum computing help doctors find cancer earlier, treat it more precisely, and reduce human suffering?

On August 11, 2026, USC ISI published details of Kalev's latest work combining quantum computing with artificial intelligence for medical image segmentation—a critical step in radiation treatment planning.

The clinical problem: segmentation

Before AI can help physicians decide how to treat a patient, it must perform image segmentation—tracing a tumor's exact outline pixel by pixel, separating cancerous tissue from healthy tissue.

Those boundaries can make all the difference. The more accurately doctors can see where a tumor begins and ends, the more precisely they can target radiation, plan surgery, and measure whether treatment is working.

Kalev suspected that quantum computing could help AI produce sharper boundaries—especially given the data constraints of medical imaging.

QuFeX and Qu-Net

Kalev teamed with his former student Naman Jain, who earned a master's degree in quantum information science in 2025. Together they created Quantum Feature Extraction (QuFeX), a quantum module designed to strengthen existing AI systems rather than replace them.

Their work was published in the journal Quantum Science and Technology.

Jain explained the insight: "The real insight was the data problem. In medicine, datasets are really small and often of very low quality. Traditional AI methods need massive amounts of data to be reliable. We wanted to see if quantum could close that gap."

After developing QuFeX, the team integrated it into U-Net, one of the most widely used AI systems for analyzing medical images. The result was Qu-Net, a hybrid quantum-classical system.

Doing more with less

On several image segmentation benchmarks including medical datasets, Qu-Net outperformed a leading classical AI model while using roughly one sixth as many parameters:

• Qu-Net: approximately 250,000 parameters
• Classical U-Net: approximately 1.5 million parameters
• Performance improvement: roughly 7% over the classical baseline

Kalev noted: "What surprised us the most wasn't just that it beat the classical baseline by roughly 7%, but that it did so using only about 250,000 parameters, compared to 1.5 million parameters for U-Net."

Jain added: "The game we're playing isn't just about whether quantum is fast. It's whether it's also efficient."

More importantly, Qu-Net produced more accurate boundaries around suspicious tissue, giving physicians a clearer map of what should be treated and what should be left alone.

Moving toward clinical care

Kalev has begun working with physicians at the Keck School of Medicine of USC. The collaboration pairs him with Dr. Eric Chang, chair of the Department of Radiation Oncology at Keck.

They are exploring whether quantum machine learning can speed one of the most critical steps in radiation treatment planning: outlining tumors and healthy organs before treatment begins.

Chang stated: "The most exciting aspect of this research is that it could begin to tackle the laborious processes involved in delivering radiation therapy, such as manual image segmentation of a patient's organs and tumors."

The researchers soon plan to test their technology using simulated scans and real patient images. Kalev's long-term goal is shrinking the time to create a personalized radiation treatment plan from days to a single clinic visit.

Limitations and next steps

The work remains in early stages. Qu-Net's advantages have been demonstrated on benchmarks and selected medical datasets—not yet in prospective clinical trials with patient outcomes as endpoints. Quantum hardware access and integration into hospital imaging pipelines present additional engineering hurdles.

Kalev emphasized the motivation: "For me, projects like QuFeX and Qu-Net aren't about proving a theoretical physics concept. They are about the tangible, real-world impact of quantum technology on human lives."

Sources

• USC Information Sciences Institute — USC Scientists Are Using Quantum Computing to Rethink Cancer Detection (August 11, 2026)
• Quantum Science and Technology — QuFeX: Quantum Feature Extraction Module for Hybrid Quantum-Classical Deep Neural Networks (2026)