This study systematically examines the impact of training database size and the generalizability of deep learning models for synthetic medical image generation. Specifically, we employ a Cycle-Consistency Generative Adversarial Network (CycleGAN) with softly paired data to synthesize kilovoltage computed tomography (kVCT) images from megavoltage computed tomography (MVCT) scans. Unlike previous works, which were constrained by limited data availability, our study uses an extensive database comprising 4,000 patient CT scans, an order of magnitude larger than prior research, allowing for a more rigorous assessment of database size in medical image translation. We quantitatively evaluate the fidelity of the generated synthetic images using established image similarity metrics, including Mean Absolute Error (MAE) and Structural Similarity Index Measure (SSIM). Beyond assessing image quality, we investigate the model's capacity for generalization by analyzing its performance across diverse patient subgroups, considering factors such as sex, age, and anatomical region. This approach enables a more granular understanding of how dataset composition influences model robustness.

Background: This work aimed to determine the optimum VOLO^TM Ultra algorithm parameters for tomotherapy treatments. Methods: 1056 treatment plans were generated with VOLO^TM Ultra for 36 patients and six anatomical locations. The impact of varying four parameters was studied: the accelerated treatment (AT), leaf open/close time (LOT) cutoff, normal tissue objective (NTO) weight, and number of iterations. The beam-on time and dosimetric metrics were quantified for the target volumes and organs at risk (OARs). Delivery quality assurance measurements were obtained for 36 plans to assess the delivery accuracy. Results: The mean beam-on time for the helical tomotherapy and TomoDirect (TD) plans decreased by 26.6 ± 2.8% and 17.4 ± 4.3%, respectively, when the accelerated treatment parameter was increased from 0 to 10, at the expense of the planning target volume (PTV) coverage (2% lower D_98%) and OAR dose (up to 15% increase). For TD plans, it seems preferable to systematically use an AT value of 10. Increasing the number of iterations beyond six seems unnecessary. In this study, an NTO weight of approximately 10 appears to be ideal and eliminates the need to use rings in the treatment plan. Finally, no correlation was found between the leaf open/close time cutoff and the delivery accuracy, while a leaf open/close cutoff of 60 ms seemed to degrade dosimetry quality. Conclusion: Optimal values for the AT, LOT cutoff, NTO weight, and number of optimization rounds were identified and should help improve the management of patients whose tomotherapy treatments are planned with VOLO^TM Ultra.

Background: The main advantages of stereotactic radiotherapy (SRT) are to delay whole-brain radiotherapy (WBRT) and to deliver ablative doses. Despite this efficacy, the risk of distant brain metastases (BM) one year after SRT ranges from 26% to 77% and 20 to 40% of patients required salvage treatment. The role and consequences of reirradiation remain unclear, particularly in terms of survival. The objective was to study overall survival (OS) and neurological death-free survival (NDFS) and to specify the prognostic factors of long-term survival. Methods: we retrospectively reviewed the data of patients treated between 2010 and 2020 with at least two courses of SRT without previous WBRT. Results: In total, 184 patients were treated for 915 BMs with two-to-six SRT sessions. Additional SRT sessions were provided for local (5.6%) or distant (94.4%) BM recurrence. The median number of BMs treated per SRT was one with a median of four BMs in total. The mean time between the two SRT sessions was 8.9 months (95%CI 7.7–10.1) and there was no significant difference in the delay between the two sessions. The 6-, 12- and 24-month NDFS rates were 97%, 82% and 52%, respectively. The 6-, 12- and 24-month OS rates were 91%, 70% and 38%, respectively. OS was statistically related to the number of SRT sessions (HR = 0.48; p < 0.01), recursive partitioning analysis (HR = 1.84; p = 0.01), salvage WBRT (HR = 0.48; p = 0.01) and brain metastasis velocity (high: HR = 13.83; p < 0.01; intermediate: HR = 4.93; p < 0.01). Conclusions: Lung cancer and melanoma were associated with a lower NDFS compared to breast cancer. A low KPS, a low number of SRT sessions, synchronous extracerebral metastases, synchronous BMs, extracerebral progression at SRT1, a high BMV grade, no WBRT and local recurrence were also associated with a lower NDFS. A high KPS at SRT1 and low BMV grade are prognostic factors for better OS, regardless of the number of BM recurrence events.