Author: Amar Kishan, Jun Lian, Yunkui Pang, Jonathan Pham, X. Sharon Qi, Michael Steinberg, Luca F Valle, Pew-Thian Yap 👨🔬
Affiliation: Department of Radiation Oncology, University of California, Los Angeles, University of North Carolina at Chapel Hill 🌍
Purpose: Stereotactic body radiotherapy (SBRT) is a highly effective treatment for prostate cancer, yet predicting genitourinary (GU) toxicity has primarily relied on planned dosimetry. This study investigates the feasibility of using delivered dosimetry from each treatment fraction for early prediction of acute GU toxicity, enabling timely adaptive treatment strategies to enhance patient outcomes.
Methods: A cohort of 78 prostate cancer patients from the MIRAGE clinical trial (NCT04384770) who received MR-guided SBRT (40 Gy in 5 fractions) on a 0.35T MR-Linac was analyzed. The dose-volume-histogram (DVH, binned in each 10 cCy interval) of the original plan and the delivered dose for each fraction, was pooled. Acute GU toxicity was assessed using CTCAE criteria, with patients categorized into significant toxicity (Grade ≥ 2, 24.4%) and non-significant toxicity (75.6%). A unidirectional Long Short-Term Memory (LSTM) model, a specialized Recurrent Neural Network (RNN), was constructed to correlate the dosimetry with toxicity outcomes. Lasso regression was applied to identify the most relevant fraction number and predictive dosimetric features for the toxicity model. Predictive accuracy was assessed using Area-Under-the-Curve (AUC).
Results: Including dosimetric data from all five fractions significantly improved the prediction accuracy for GU toxicity (AUC: 0.62±0.04), compared to planning dosimetry alone (AUC: 0.56±0.03, p=0.018). Remarkably, combining planning dosimetry with data from the first two fractions significantly enhanced prediction accuracy (AUC: 0.74±0.03, p=6e-4). Conversely, including dosimetry from fractions 4 and 5 reduced predictive accuracy compared to fractions 1 and 2 (AUC: 0.54±0.04, p=7e-3). Key predictive dosimetric features included urethra 41.5Gy volume, urethra 41.2Gy volume, and trigone 39.8Gy volume.
Conclusion: This study demonstrates the improvement in predicting GU toxicity by combining plan and delivery dosimetry for prostate SBRT treatment. The findings emphasize the importance of early fraction data for toxicity modeling, offering a pathway for real-time treatment adaptation and improved clinical outcomes.