Author: Wenbo Gu, Chenhui Qiu, Ke Sheng, Liyan Sun, Weiyuan Sun, Lei Xing 👨🔬
Affiliation: Department of Radiation Oncology, Stanford University, Department of Radiology, Stanford University, University of Pennsylvania, Department of Radiation Oncology, University of California, San Francisco, Department of Radiation Oncology, Stanford University, 🌍
Purpose:
The small animal radio-neuromodulation platform developed in our previous work utilized focused kV x-ray beams rotating and translating in predefined trajectories to irradiate small, mm size targets in rat brain. A robust inverse planning approach was developed to mitigate positioning uncertainties caused by robotic arm motion and animal immobilization, while maintaining superior target homogeneity. However, the plan quality is limited by beam directions, which is manually selected and designed. We aim to develop an integrated robust inverse planning method to optimize both directions and intensities of focused beams in 3D spherical space.
Methods:
Focused kV x-ray beams were generated using a polycapillary x-ray lens to achieve small (<0.3 mm) foci. Their directions and positions were controlled by robotic arm motion. The candidate beams included 3225 beams in a rat brain, evenly distributed within half of 3D spherical space, with 5-degree angular interval and 0.1 mm step-size displacement.
The dose matrices of all candidate beams were efficiently produced by a novel analytical dose calculation method based on 3D implicit neural representation model that built a dose kernel library.
A stochastic programming-based integrated robust inverse planning method directly selected beams and incorporated uncertainties into the planning process. An objective function was established to penalize the quadratic difference between the delivered and prescribed (20 Gy) doses to target, while minimizing the dose to rat skull, therefore determining both directions and intensities of those selected beams for optimal dosimetry and robustness.
Results:
The proposed method improved homogeneity from 18.3% to 15.7% and bandwidth at D90 from 3.3 Gy to 3.1 Gy compared to previous method, the dose calculation time of candidate beams distinctly reduced from ~48 hours to ~25 mins.
Conclusion:
The proposed method significantly reduced dose calculation time and slightly improved dose uniformity and plan robustness, which are crucial for radio-neuromodulation.