Bridging speed and accuracy in radiation therapy QA
Led by Professor Fu Jin, the study addresses a critical challenge in radiation therapy: balancing the computational speed and accuracy of EPID-based dose verification. EPID has emerged as a key tool for real-time in vivo dose verification. However, MC simulation, long considered the “gold standard” for dose calculation, faces a dilemma: increasing the number of simulated particles ensures higher accuracy, but at the cost of significantly longer calculation times, while decreasing the number of particles introduces disturbing noise that compromises the reliability of the results.
Built-in MC-DL technology
To address this challenge, the team combined the GPU-accelerated MC ARCHER code with the SUNet neural network—a sophisticated deep learning architecture specialized in denoising. Using IMRT lung cancer cases, they first generated noisy EPID transmission dose data with four different particle numbers (1×106, 1×107, 1×108, 1×109) via ARCHER. SUNet was then trained to denoise the low particle number data, with the high-fidelity 1×10⁹ particle data set serving as the gold standard reference for supervision.
Remarkable results: Speed and accuracy achieved
The embedded MC-DL framework demonstrated excellent performance in both computational speed and dosimetric accuracy. When processing the initially noisy 1×106‑particle data, SUNet denoising improved the structural similarity index (SSIM) from 0.61 to 0.95 and increased the gamma-pass rate (GPR) from 48.47% to 89.10%. For the 1×107-particle dataset representing an optimal trade-off, the denoised results achieved SSIM 0.96 and GPR 94.35%, while the 1×108-particle case reached 99.55% GPR after processing. The denoising step itself required only 0.13–0.16 s, reducing the total computation time to 1.88 s for the 1×107 particle level and to 8.76 s for the 1×108 particle level. Denoised images showed markedly reduced graininess, with smooth dose profiles that preserved clinically relevant features—confirming the practical viability of this approach for effective QA in radiotherapy.
Strengthening clinical practice and future research
This advance is particularly important for online ART, where rapid dose verification is essential to minimize patient discomfort and mitigate anatomical variations during treatment. The method offers a flexible solution: 1×107 particles achieve the optimal balance between speed and accuracy for time-sensitive scenarios, while 1×108 particles provide higher accuracy for demanding cases.
“By combining the accuracy of Monte Carlo simulation with the computational efficiency of deep learning, we have developed a practical solution that addresses the critical clinical need for fast and reliable patient-specific quality assurance,” said Professor Fu Jin. “This technology not only enhances existing radiation therapy workflows, but also lays the foundation for advanced applications such as 3D dose reconstruction and broader application in various anatomical locations.”
The team plans to extend the model to other treatment sites, further optimize the SUNet architecture, and explore additional neural network approaches to improve dose prediction capabilities.
