S1748
Physics - Dose prediction/calculation, optimisation and applications for particle therapy planning
ESTRO 2026
architecture was developed for CNN and DECT-based estimation (DECT-CNN). For calibration and CNN training, a simple geometrical phantom with known elemental composition was employed to ensure ground-truth accuracy. The models were then evaluated using the JM-103 voxel phantom, which reproduces realistic human anatomy and consists of known tissue compositions. To simulate imaging uncertainties, Gaussian random noise with standard deviation σ and systematic CT-number offsets Δ HU were added to the CT images. The resulting change in mean absolute error of SPR ( Δ MAE_SPR) was used to quantify sensitivity to random and systematic errors, represented by Δ MAE_SPR / σ and Δ MAE_SPR / Δ HU, respectively. Results: The sensitivity to random CT-number errors ( Δ MAE_SPR / σ ) was 0.030 %/HU, 0.148 %/HU, and 0.017 %/HU for SECT, DECT-HS, and DECT-CNN, respectively, indicating that DECT-CNN achieved the highest robustness to random noise. The sensitivity to systematic errors ( Δ MAE_SPR / Δ HU) was 0.064 %/HU, 0.091 %/HU, and 0.069 %/HU, indicating that SECT and DECT-CNN were less affected by systematic CT- number shifts than DECT-HS. Conclusion: The conventional DECT-HS method may yield significant SPR errors when CT-number uncertainties are present, potentially limiting its clinical reliability. The proposed CNN-based approach demonstrated strong robustness to both random and systematic CT- number errors, enabling accurate and stable SPR estimation even under imperfect imaging conditions. These findings suggest that integrating CNN-based DECT processing could enhance the precision and reliability of dose calculation of proton therapy. References: 1) Schneider et al: The calibration of CT Hounsfield units for radiotherapy treatment planning. Phys Med Biol 41 (1): 111-124, 19962) Saito et al: Simplified derivation of stopping power ratio in the human body from dual-energy CT data. Med Phys 44 (8): 4179- 4187, 20173) Hünemohr et al: Experimental verification of ion stopping power prediction from dual energy CT data in tissue surrogates. Phys Med Biol 59 (1): 83-96, 2014 Keywords: stopping power, neural network, dual- energy CT
factors (34.7%) contributed.
Conclusion: Plan adaptation is frequent and site-dependent in both proton and C-ion therapy, with the highest rates in H&N, gynecological, and pelvic sites. Systematic image-based anatomical monitoring and adaptive replanning are fundamental to ensure treatment accuracy in PT. Keywords: adaptive, planning workload, multi-year analysis Sensitivity analysis for calculating stopping power ratios using a convolutional neural network and single- and dual-energy computed tomography Takayuki Kanai 1 , Izobelle Echaluse 2 , Weishan Chang 2 , Yuhei Kikkawa 1 , Yaichiro Hashimoto 1 1 Department of Radiation Oncology, Tokyo Women's Medical University, Tokyo, Japan. 2 Department of Radiological Sciences, Tokyo Metropolitan University, Tokyo, Japan Digital Poster 2976 Purpose/Objective: Accurate estimation of the stopping-power ratio (SPR) is essential in proton therapy dose calculation, as even a few percent error in SPR can result in millimeter- level range deviations in patient dose delivery. While dual-energy CT (DECT) has demonstrated improved SPR accuracy compared to conventional single-energy CT (SECT), several studies have reported that DECT- based analytical methods are susceptible to CT- number errors. Such sensitivity can lead to considerable uncertainties in clinical dose calculations. This study aimed to investigate whether a convolutional neural network (CNN)–based DECT approach can achieve robust SPR estimation against both random and systematic CT-number errors. Material/Methods: Three different SPR estimation methods were compared. The SECT-based method followed the model proposed by Schneider et al., representing the current clinical standard. The DECT-based analytical method followed Hünemohr and Saito’s formulation (DECT-HS), which is widely used in the DECT-based approach. In addition, an early-fusion U-Net
Mini-Oral 3050
Photon-counting CT for robust HLUT calibration and artifact reduction in proton therapy planning Chiara Radice 1,2 , Andrea Bresolin 2 , Pasqualina Gallo 2 , Daniel Maneval 3 , Abdallah Alshatali 4 , Laura Breschi 2 , Francesco La Fauci 2 , Nicola Lambri 1,2 , Francesca
Made with FlippingBook - Share PDF online