Simulation-free spine palliative radiotherapy enabled by AI-adapted diagnostic CT

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiation Oncology
• Radiotherapy Planning

Background:

• Traditional radiotherapy planning requires a simulation CT (sCT), causing treatment delays.
• Diagnostic CT (dCT) is often available earlier but has geometric and dosimetric inaccuracies for planning.
• Timely treatment is crucial in spinal palliative care to manage symptoms and prevent neurological decline.

Purpose of the Study:

• To develop and validate an AI-based method (AI-pCT) for transforming dCT into sCT-equivalent images.
• To enable a clinically feasible simulation-free workflow for spinal palliative radiotherapy.

Main Methods:

• Two neural networks were trained on paired dCT-sCT images to correct spine position and body contour.
• The AI-pCT method was evaluated on internal test and external academic medical center cohorts.
• Dosimetric accuracy was assessed by comparing dose-volume histogram (DVH) endpoints and RMS error between dCT, AI-pCT, and sCT plans.

Main Results:

• AI-pCT significantly reduced geometric and dosimetric errors compared to dCT in the safety net cohort (e.g., Dmean error reduced from 2.0% to 0.57%).
• Physician plan-quality ratings improved from "Acceptable" to "Good-Perfect," and clinical goal achievement increased from 37.5% to 100%.
• Statistically significant gains were also observed in the external academic medical center cohort.

Conclusions:

• AI-pCT achieves sCT-level fidelity, enabling simulation-free planning for spinal palliative radiotherapy.
• This AI-driven approach can potentially reduce treatment delays and improve access to care, especially in resource-limited settings.