Abstract Carbon ion radiotherapy (CIRT) has evolved into an established clinical modality over the past three decades. Owing to its unique physical and biological characteristics, CIRT has demonstrated favorable outcomes for selected malignancies characterized by radioresistance, proximity to critical organs, locally advanced disease requiring safe dose escalation, and recurrent tumors requiring re‐irradiation. This review summarizes the physical and radiobiological rationale for CIRT, the evolution of international clinical practice and the clinical relative biological effectiveness models and dose prescription systems currently used in Asia and Europe. We discuss the differences between the modified microdosimetric kinetic model and Local Effect Model frameworks and their implications for interpreting and applying international clinical experience. We also describe the biological dose calculation framework adopted by Mayo Clinic Florida (MCF) to integrate established international CIRT experience into the first U.S. program. Published estimates of CIRT eligibility and need in the United States are reviewed to provide context for the anticipated U.S. patient population. This population includes many clinically complex disease sites characterized by proximity to critical organs, respiratory motion, anatomical variation, and the potential need for adaptive treatment. These clinical characteristics define important requirements for the first U.S. CIRT program and provide the foundation for the design philosophy and technical implementation described in the companion Part II article.
Abstract Purpose Workforce shortages and increasing clinical demands in Radiation Oncology and Medical Physics present challenges for maintaining high‐quality residency training. This work presents a conceptual framework to evaluate the potential role of artificial intelligence (AI) in improving efficiency, scalability, and quality of residency education. Methods Current teaching and learning modalities were categorized, and their relative contributions to training were estimated. Potential AI applications were mapped to these modalities, and the proportion of training that could be supported by AI was assessed. A simple conceptual model incorporating training time, AI‐deliverable components (i.e., those potentially supported through direct interaction with AI systems), and dependence on direct human teaching was used to illustrate the potential impact on instructional effort. The potential effect of AI on educational quality was qualitatively evaluated, as well as limitations of AI for routine educational integration. Results AI applications were identified across multiple domains, including workflow‐integrated feedback, simulation‐based training, adaptive learning, and automated assessment. The largest potential impact was estimated in workflow‐integrated, simulation‐based, and self‐directed learning environments. While reductions in direct human teaching effort were estimated to be modest, AI‐enabled approaches were estimated to achieve substantial improvements in training quality, including increased feedback frequency, enhanced support for deliberate practice, and greater exposure to rare or high‐risk scenarios. Conclusions AI has the potential to enhance residency training by improving consistency, scalability, and educational quality. In the context of workforce constraints, the primary benefit of AI may be to augment training efficiency and reduce reliance on limited faculty resources while preserving the essential role of clinical experience.
Abstract Purpose/Objective We report our institutional clinical experience of treatment planning, dose evaluation, and delivery via the same‐day 3D MLC‐based crossfire method for spatially fractionated radiotherapy (SFRT) in conjunction with combination therapy. Additionally, we aimed to model our clinical outcome data to provide clinical insights into treatment planning using this SFRT method for unresectable bulky tumors. Materials/Methods A total of 137 large tumors in 131 patients with different histopathologies were treated with the 3D MLC‐based SFRT method using MLC‐fitting for a 1 cm cylindrical high‐dose and a 2 cm center‐to‐center distance at the isocenter. Physical and biological effective dose metrics for the SFRT target were reported, as were the maximum EQD2 to nearby critical organs. Clinical follow‐up data included tumor local control evaluated on post‐treatment CT scans, pain relief, toxicity profiles, and overall survival (OS). Tumor control probability (TCP) and pain relief probability were modeled using multivariable logistic regression, while normal tissue complication probability (NTCP) was modeled using univariable logistic regression. Lastly, OS was modeled with a univariable Cox hazard model. Results This cohort had a median tumor volume of 477.55 cc (range, 21.47–7879.87 cc). The patient cohort was predominantly male (61.8%), with most treated sites in the abdominal‐pelvic area (37.2%). Furthermore, the most common SFRT prescription was 15 Gy in 1 fraction (85.4%), and most tumors received follow‐up treatment (126/137) with palliative intent (75/126). The median SFRT dose metric values for D5%, D10%, D50%, Dmean, D90%, D95%, EUD, and PVDR are 13.2 Gy, 11.8 Gy, 6.9 Gy, 7.3 Gy, 3.5 Gy, 2.4 Gy, 5.3 Gy, and 3.4. The resulting median target BED metrics, BED5%, BEDmean, and BED95% was 71.5, 54.6, and 44.4 Gy. A TCP rate of 76.5% (75/98), and a pain relief rate of 72.0% (59/82) was found. Using multivariable logistic regression, none of these dose metrics were statistically significant with TCP or pain relief outcomes. In total, 95 SFRT patients were available for toxicity assessment, with 65 (68.4%) reporting no toxicities and 7 (7.4%) reporting grade 3+ events. Skin toxicity and xerostomia were statistically significantly correlated with maximum EQD2 values to the skin ( p = 0.022) and parotid glands ( p = 0.038). The Cox hazard model showed that increases in BED metrics were statistically significantly associated with longer OS; however, increases in SFRT dose metrics were not. We reported OS rates of 51.8%, 36.3%, and 30.3% at 6, 12, and 18 months, respectively. Moreover, we found differences in OS between patients who received a combination palliative treatment to those who received a curative treatment ( p = 0.002). Conclusion Here, we report our single‐institution results using the 3D‐MLC‐based crossfire technique for SFRT treatment and the resulting outcome models. Further refinement of this predictive model for SFRT in a larger, more diverse cohort is ongoing to validate our findings.
Abstract Purpose : During lung cancer radiotherapy, inter‐fractional anatomical changes may cause the delivered dose to deviate from the primary plan, affecting tumor control and increasing the risk of normal tissue damage. This study aims to evaluate the impact of anatomical changes on the dosimetric and radiobiological performance of simulated proton arc therapy (sPAT) versus intensity‐modulated proton therapy (IMPT) for lung cancer patients under patient‐specific anatomical changes. Methods : Data from 21 lung cancer patients were retrospectively analyzed. The prescribed dose for all patients was 6000 cGy (relative biological effectiveness, RBE). Each patient with a primary planning computed tomography (pCT) and a mid‐treatment adaptive‐planning CT (aCT) acquired at a median of 29 days into treatment due to notable tumor volume changes. Primary plans for sPAT and IMPT were designed on the pCT. After rigid registration of the pCT and aCT, non‐adaptive plans (sPAT p‐a , IMPT p‐a ) were generated by directly recalculating the doses of the primary plans on the aCT. The comparison evaluated the dosimetric impact of anatomical changes by comparing non‐adaptive plans (sPAT p‐a vs. IMPT p‐a ). Dose‐volume histogram (DVH) parameters for targets and organs at risk (OARs) and tumor control probability (TCP) and normal tissue complication probability (NTCP) were evaluated. Results : For the primary plans, sPAT and IMPT achieved similar target coverage, while the doses to OARs were generally lower in sPAT p . Under anatomical changes, both non‐adaptive plans exhibited a decrease in target coverage, sPAT p‐a achieved a D 95% of 5567.57 cGy[RBE] compared with 5423.33 cGy[RBE] for IMPT p‐a ( p < 0.001), indicating that sPAT better preserved target coverage. Compared with the primary plans, sPAT p‐a and IMPT p‐a exhibited TCP reductions of 5.83% and 8.07%, respectively. Conclusions : Compared with IMPT, sPAT exhibits enhanced robustness against inter‐fractional anatomical changes occurring in lung‐cancer radiotherapy.
Abstract Background Clinical medical physicists in a radiotherapy setting are aware of the importance of maintaining linac collimators with high positional accuracy, especially when used to form junctions; however, there is a lack of quality control options available. This is especially true when trying to improve jaw matching in junctions or when trying to maintain accurate jaws across gantry angles other than 0 degrees. Purpose In this work, we address this quality assurance niche by developing a jaw calibration optimization program, which defines jaw calibration positions to minimize jaw positional errors and junction matching errors over customizable gantry and collimator angles. Methods A jaw calibration optimization program was developed and evaluated on two Varian TrueBeam linacs. Half beam blocked MV EPID images were acquired to characterize jaw positions and clinically relevant breast junction errors across a set of gantry and collimator angles. Collimator encoder values were correlated with measured jaw positions, allowing the jaws to be set precisely during calibration. The Jaw calibration program analyzed DICOM images, modelled encoder‐position relationships, and optimized the calibration origins for all four jaws. The jaw calibration was performed and then verified by redoing the analysis with the new jaw settings, and standard QA of graph paper jaw position, light field radiation coincidence, and jaw matching. Results The jaw calibration was found to reduce clinically relevant junction errors on both units, from a maximum of 1.5 mm to 0.3 mm on Unit A and 1.2 mm to 0.7 mm on Unit B. The sum of cold junctions was reduced from 9.5 mm to 0.8 mm and from 4.9 to 0.6 mm, respectively. Mean jaw positional errors decreased significantly for some jaws and remained constant for others. Predicted jaw positions following calibration agreed well with measurements. Conclusions Our results demonstrate the viability of this jaw calibration optimization program as a quality assurance tool. This program has been made open source and provides a unique QA solution for a common issue in centers operating linear accelerators for radiotherapy.
