Modeling ultra-high dose rate electron and proton FLASH effect with the physicochemical approach.

Objective. A physicochemical model built on the radiochemical kinetic theory was recently proposed in (Labarbeet al2020) to explain the FLASH effect. We performed extensive simulations to scrutinize its applicability for oxygen depletion studies and FLASH-related experiments involving both proton and electron beams.Approach. Using the dose and beam delivery parameters for each FLASH experiment, we numerically solved the radiochemical rate equations comprised of a set of coupled nonlinear ordinary differential equations to obtain the area under the curve (AUC) of radical concentrations.Main results. The modeled differences in AUC induced by ultra-high dose rates appeared to correlate well with the FLASH effect. (i) For the whole brain irradiation of mice performed in (Montay-Gruelet al2017), the threshold dose rate values for memory preservation coincided with those at which AUC started to decrease much less rapidly. (ii) For the proton pencil beam scanning FLASH of (Cunninghamet al2021), we found linear correlations between radicals' AUC and the biological endpoints: TGF-ß1, leg contracture and plasma level of cytokine IL-6. (iii) Compatible with the findings of the proton FLASH experiment in (Kimet al2021), we found that radicals' AUC at the entrance and mid-Spread-Out Bragg peak regions were highly similar. In addition, our model also predicted ratios of oxygen depletionG-values between normal and UHDR irradiation similar to those observed in (Caoet al2021) and (El Khatibet al2022).Significance. Collectively, our results suggest that the normal tissue sparing conferred by UHDR irradiation may be due to the lower degree of exposure to peroxyl and superoxide radicals. We also found that the differential effect of dose rate on the radicals' AUC was less pronounced at lower initial oxygen levels, a trait that appears to align with the FLASH differential effect on normal versus tumor tissues.

Deep learning for automatic target volume segmentation in radiation therapy: a review.

Deep learning, a new branch of machine learning algorithm, has emerged as a fast growing trend in medical imaging and become the state-of-the-art method in various clinical applications such as Radiology, Histo-pathology and Radiation Oncology. Specifically in radiation oncology, deep learning has shown its power in performing automatic segmentation tasks in radiation therapy for Organs-At-Risks (OAR), given its potential in improving the efficiency of OAR contouring and reducing the inter- and intra-observer variabilities. The similar interests were shared for target volume segmentation, an essential step of radiation therapy treatment planning, where the gross tumor volume is defined and microscopic spread is encompassed. The deep learning-based automatic segmentation method has recently been expanded into target volume automatic segmentation. In this paper, the authors summarized the major deep learning architectures of supervised learning fashion related to target volume segmentation, reviewed the mechanism of each infrastructure, surveyed the use of these models in various imaging domains (including Computational Tomography with and without contrast, Magnetic Resonant Imaging and Positron Emission Tomography) and multiple clinical sites, and compared the performance of different models using standard geometric evaluation metrics. The paper concluded with a discussion of open challenges and potential paths of future research in target volume automatic segmentation and how it may benefit the clinical practice.