This study aimed to integrate Monte Carlo (MC) simulation with deep learning (DL)-based denoising techniques to achieve fast and accurate prediction of high-quality electronic portal imaging device (EPID) transmission dose (TD) for patient-specific quality assurance (PSQA). A total of 100 lung cases were used to obtain the noisy EPID TD by the ARCHER MC code under four kinds of particle numbers ( 1× 10^6 , 1× 10^7 , 1× 10^8 and 1× 10^9 ), and the original EPID TD was denoised by the SUNet neural network. The denoised EPID TD was assessed both qualitatively and quantitatively using the structural similarity (SSIM), peak signal-to-noise ratio (PSNR), and gamma passing rate (GPR) with respect to 1× 10^9 as a reference. The computation times for both the MC simulation and DL-based denoising were recorded. As the number of particles increased, both the quality of the noisy EPID TD and computation time increased significantly ( 1× 10^6 : 1.12 s, 1× 10^7 : 1.72 s, 1× 10^8 : 8.62 s, and 1× 10^9 : 73.89 s). In contrast, the DL-based denoising time remained at 0.13 - 0.16 s. The denoised EPID TD shows a smoother visual appearance and profile curves, but differences between 1× 10^6 and 1× 10^9 still remain. SSIM improves from 0.61 to 0.95 for 1× 10^6 , 0.70 to 0.96 for 1× 10^7 , and 0.90 to 0.97 for 1× 10^8 . PSNR increases by > 20 1× 10^6 and 1× 10^7 , and > 10 1× 10^8 . GPR improves from 48.47 1× 10^6 , 61.04 1× 10^7 , and 91.88 1× 10^8 . The method that combines MC simulation with DL-based denoising for EPID TD generation can accelerate TD prediction and maintain high accuracy, offering a promising solution for efficient PSQA.

Evolutionary multitasking optimization, where the auxiliary task transfers knowledge to assist the main task, has been recognized in solving constrained multi-objective optimization problems. However, the auxiliary task of most algorithms encounters difficulties in transferring useful knowledge due to quantities of decision variables and constraints in volumetric modulated arc therapy planning. To tackle this issue, an effective auxiliary task is tailored to this problem, which features a small number of decision variables and relaxed constraints. The auxiliary task is derived from cloning the original problem. In particular, the constraint information of the original problem is utilized to reduce decision variables, and the constraints are relaxed accordingly. With the proposed auxiliary task, a multitask-based constrained evolutionary algorithm is developed for volumetric modulated arc therapy planning(MTCEA/VMAT). Experimental results on six clinical cases exhibit the superior performance of the proposed algorithm against representative algorithms.

Objective.Intensity-modulated brachytherapy (IMBT) is an innovative technique aimed at achieving anisotropic dose distributions in brachytherapy. This study develops a fast dosimetric optimization method specifically for IMBT plans in cervical cancer.Approach.ARCHER-IMBT was validated against TOPAS in both water phantoms and clinical geometries. Optimization was performed for six intracavitary (IC-BT) cases and one intracavitary/interstitial (IC/IS-BT) case, comparing 50 kVp x-ray and Ir-192 sources. The study also explored the potential of IMBT to achieve comparable dosimetry to IC/IS-BT using only intracavitary applicators. Furthermore, a stochastic uncertainty analysis (200 Monte Carlo scenarios) was conducted to evaluate plan robustness against positional (0.3 mm) and angular (0.2°) perturbations.Main results.ARCHER-IMBT achieved speedup factors exceeding 50× for water phantoms and 350× for clinical cases, with gamma passing rates >98%. The entire optimization process was completed within one minute. Compared to IC-BT, IMBT plans reduced bladder and rectumD2ccby 3.1% and 15.1% for Ir-192, and by 23.4% and 22.8% for 50 kVp x-rays, respectively. In the IC/IS-BT case, IMBT plans achieved comparable target coverage while potentially eliminating the need for invasive needles. However, uncertainty analysis revealed that the 50 kVp source is highly sensitive to sub-millimeter translational errors (0.3 mm) due to its steep dose gradients, whereas Ir-192 exhibited greater robustness.Significance.This study demonstrates a computationally efficient IMBT optimization platform. The findings highlight the dosimetric benefits of IMBT and its potential to simplify complex IC/IS-BT procedures, while underscoring the stringent mechanical precision required for clinical implementation.

The efficacy of Boron Neutron Capture Therapy (BNCT) critically depends on the characteristics of the neutron beam shaped by the Beam Shaping Assembly (BSA). However, systematic studies linking BSA design parameters to beam characteristics and clinical dosimetry remain limited. This study employed Monte Carlo (MC) simulations to generate Phase Space Files (PSFs) for various BSA configurations, which were then used as neutron sources for BNCT treatment planning. Analysis of the PSFs revealed that moderator material primarily influenced the neutron energy spectrum, while collimator length affected neutron flux density, directionality, and photon yield. Specifically, increasing the collimator length by 10 cm reduced neutron flux density and photon yield by approximately 32% and 17.9%, respectively. Although longer collimators improved Gross Tumor Volume (GTV) dose via enhanced beam collimation, they also prolonged treatment length by about 50% and increased external photon contribution to the total dose. Additionally, higher high-energy neutron content in the PSF elevated fast neutron dose to the skin. By establishing a clear link from BSA design to PSF characteristics and clinical outcomes, this study provides a physical and methodological basis for personalized BNCT planning.

To investigate the best combination of parameters by changing the collimator and increment in the volumetric modulated arc therapy after modified radical mastectomy for breast cancer. Ten patients with left breast cancer and ten patients with right breast cancer who underwent modified radical mastectomy were selected in our research cohort. Treatment plans were established utilizing the 3-arc beam configuration spanning 240 degrees. The collimator angle was systematically adjusted in increments of 10 degrees, ranging from - 90 to 90 degrees. Additionally, increment values of 10, 20, 30, and 40 were employed for each corresponding collimator angle. All other planning parameters remained constant across the plans. Finally the impact of different values was analyzed in terms of plan quality and execution efficiency. Regarding the dose distribution within the target and the ipsilateral lung, the optimal collimator angles were observed to range from - 60 to 30 degrees for the left breast and from - 30 to 60 degrees for the right breast. As for the increment value, 10 yielded the optimal outcome, 40 resulted in the least desirable outcome, and 20 and 30 demonstrated a balanced effect between 10 and 40. In terms of execution efficiency and treatment complexity, the impact of the collimator angle on the results exhibits symmetric distribution, and the most favorable collimator angle approaches 0 degrees. Additionally, with a gradual increase in the collimator angle, the execution efficiency diminishes while the complexity correspondingly increases. In the context of treatment planning following modified radical mastectomy for left and right breast cancer, optimal collimator angles fall within the ± 30-degree range, while increment values of 20 and 30 degrees yielded the best overall outcomes, ensuring the attainment of the prescribed dose while accounting for execution efficiency and treatment complexity.

To evaluate the precision of automated segmentation facilitated by deep learning (DL) and dose calculation in adaptive radiotherapy (ART) for nasopharyngeal cancer (NPC), leveraging synthetic CT (sCT) images derived from cone-beam CT (CBCT) scans on a conventional C-arm linac. Sixteen NPC patients undergoing a two-phase offline ART were analyzed retrospectively. The initial (pCT1) and adaptive (pCT2) CT scans served as gold standard alongside weekly acquired CBCT scans. Patient data, including manually delineated contours and dose information, were imported into ArcherQA. Using a cycle-consistent generative adversarial network (cycle-GAN) trained on an independent dataset, sCT images (sCT1, sCT4, sCT4*) were generated from weekly CBCT scans (CBCT1, CBCT4, CBCT4) paired with corresponding planning CTs (pCT1, pCT1, pCT2). Auto-segmentation was performed on sCTs, followed by GPU-accelerated Monte Carlo dose recalculation. Auto-segmentation accuracy was assessed via Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (HD95). Dose calculation fidelity on sCTs was evaluated using dose-volume parameters. Dosimetric consistency between recalculated sCT and pCT plans was analyzed via Spearman’s correlation, while volumetric changes were concurrently evaluated to quantify anatomical variations. Most anatomical structures demonstrated high pCT-sCT agreement, with mean values of DSC > 0.85 and HD95 < 5.10 mm. Notable exceptions included the primary Gross Tumor Volume (GTVp) in the pCT2-sCT4 comparison (DSC: 0.75, HD95: 6.03 mm), involved lymph node (GTVn) showing lower agreement (DSC: 0.43, HD95: 16.42 mm), and submandibular glands with moderate agreement (DSC: 0.64–0.73, HD95: 4.45–5.66 mm). Dosimetric analysis revealed the largest mean differences in GTVn D99: -1.44 Gy (95

Objective. Currently, superficial x-ray radiotherapy does not take advantage of modern treatment planning technologies. To address the problem, a treatment planning system for superficial x-ray radiotherapy has been developed using a database generated through Monte Carlo simulations. The system, called SXRTDose, can be used to optimize irradiation strategies by adjusting energy, filtration, and applicator aiming to deliver a planned dose to the target volume while minimizing radiation risk to surrounding normal tissues.Approach. TOPAS Monte Carlo code was used for establishing the dosimetric database by modeling parameters of a commercial superficial x-ray radiotherapy device and by calculating depth-dose information in a water phantom. After the radiation physics aspects have been verified, detailed Monte Carlo simulations of absorbed doses under different irradiation parameters including five skin models (representing location of the abdomen, cheek, forehead, limbs, and nose), three x-ray energies (50 kV, 70 kV, and 100 kV), corresponding filters and applicators were performed resulting in a comprehensive database. A python-based graphical user interface was developed to support the clinical application of the treatment planning system for superficial x-ray radiotherapy.Results. Compared to experimental results reported in the literature, the relative errors from water phantom simulations for the superficial x-ray radiotherapy system is acceptable. The developed treatment planning system utilizes dose-volume histograms to quantitatively evaluate the clinical applicability of various irradiation plans for skin cancer treatment. The application of the software is found to provide rapid and accurate dose guidance to clinical users in selecting optimal and alternative equipment parameters.Conclusion. The potential and feasibility of a treatment planning system for superficial x-ray radiotherapy have been evaluated, demonstrating its capability to deliver rapid, accurate, and concise dosimetry references. This enhances therapeutic guidance and treatment effectiveness, while addressing the present challenge of inadequate dosimetry support in the field of superficial radiotherapy.

This work presents an effort to extend the capabilities of the previously introduced GPU-based Monte Carlo code ARCHER for helium ion therapy. ARCHER performs helium ion transport simulations in voxelized geometry, covering kinetic energy levels up to 220 MeV/u. The physical processes are modeled using a class II condensed-history algorithm, considering ionization, energy straggling, multiple scattering, and elastic and inelastic nuclear interactions. A new nuclear-event-repeat algorithm is proposed to generate inelastic nuclear reaction products. Secondary protons, deuterons, tritons, and 3He particles are tracked, while other particles either deposit their energy locally or are ignored. The code is developed under the compute unified device architecture (CUDA) platform to improve computational efficiency. Validations are conducted by benchmarking our code against TOPAS in different phantoms. Dose distribution comparisons demonstrate strong agreement between our code and TOPAS. The mean point-by-point local relative errors in the region where the dose exceeds 10

BACKGROUND:Kilovoltage x-ray units (50-150 kVp) are widely used for skin cancer treatments, necessitating precise dosimetric characterization to ensure accurate dose delivery. This study validates the depth dose data of the VF-80/𝜇XHP (80 kV, 100 W) x-ray system for use in low-kV therapy with a contemporary detector. OBJECTIVE:This research aims to characterize the kilovoltage VF-80/𝜇XHP x-ray unit with open-ended PMMA and stainless-steel applicators of various diameters by utilizing different x-ray energy values. METHODS:Measurements included machine output reproducibility, applicator leakage and thickness effects, measurement of HVL, reference output dosimetry, and applicator-to-phantom spacing. PDD curves were obtained using various filter/kV and applicator combinations with a PTW 34013 parallel-plate ionization chamber in a solid water phantom and through MC simulations with the TOPAS Monte Carlo code in a water phantom, which were validated against literature data. RESULTS:The study evaluated around 40 x-ray beams. The PDDs were measured in a solid water phantom using a PTW 34013 ionization chamber and calculated with the TOPAS Monte Carlo code. Both the measured and computed percent depth dose data were compared to the BJR 25 supplementary. The agreement between ionization chamber measurements and MC calculations with BJR 25 supplementary was within [1.9%, 1.5%] for 50 kV x-rays and within [0.5%, 0.2%] for 70 kV x-rays. CONCLUSION:This research provides the dosimetric analysis of a kilovoltage VF-80/μXHP x-ray machine. The primary focus was on using a PTW 34013 ionization chamber to measure the percentage depth dose in a solid water phantom, along with dose calculations performed using the TOPAS Monte Carlo simulation. The findings align well with existing literature, confirming that this x-ray unit is suitable for low-kV x-ray radiotherapy applications.

Purpose: We presented a GPU-based MC framework, ARCHER-EPID, specifically designed for EPID transit dosimetry, with improving accuracy and efficiency. Methods: A comprehensive MC framework was developed to perform full radiation transport simulations through three distinct zones: a detailed linear accelerator head model, a CT-based patient/phantom geometry, and a realistic, multi-layered EPID model. To convert the simulated absorbed dose to a realistic detector signal, a dose-response correction model was implemented. The framework was validated by comparing simulations against experimental measurements for 25 IMRT fields delivered to both a solid water phantom and a anthropomorphic phantom. Agreement was quantified using Gamma analysis. Results: The GPU-accelerated ARCHER-EPID framework can complete the simulation for a complex IMRT field in about 90 seconds. A 2D correction factor lookup table is generated by parameterizing radiological thickness and effective field size to account for the EPID's energy-dependent response. The data revealed that for small fields, beam hardening is the dominant effect, while for large fields, the contribution from patient-generated scatter overwhelms this effect. The average 2D gamma passing rates (3

Objective. The primary purpose of this work is to demonstrate the feasibility of a deep convolutional neural network (dCNN) based algorithm that uses two-dimensional (2D) electronic portal imaging device (EPID) images and CT images as input to reconstruct 3D dose distributions inside the patient. Approach. To generalize dCNN training and testing data, geometric and materials models of a VitalBeam accelerator treatment head and a corresponding EPID imager were constructed in detail in the GPU-accelerated Monte Carlo dose computing software, ARCHER. The EPID imager pixel spatial resolution ranging from 1.0 mm to 8.5 mm was studied to select optimal pixel size for simulation. For purposes of training the U-Net-based dCNN, a total of 101 clinical intensive modulated radiation treatment cases-81 for training, 10 for validation, and 10 for testing-were simulated to produce comparative data of 3D dose distribution versus 2D EPID image data. The model's accuracy was evaluated by comparing its predictions with Monte Carlo dose. Main Results. Using the optimal EPID pixel size of 1.5 mm, it took about 18 min to simulate the particle transport in patient-specific CT and EPID imager per a single field. In contrast, the trained dCNN can predict 3D dose distributions in about 0.35 s. The average 3D gamma passing rates between ARCHER and predicted doses are 99.02 +/- 0.57% (3%/3 mm) and 96.85 +/- 1.22% (2%/2 mm) for accumulated fields, respectively. Dose volume histogram data suggest that the proposed dCNN 3D dose prediction algorithm is accurate in evaluating treatment goals. Significance. This study has proposed a novel deep-learning model that is accurate and rapid in predicting 3D patient dose from 2D EPID images. The computational speed is expected to facilitate clinical practice for EPID-based in-vivo patient-specific quality assurance towards adaptive radiation therapy.

The ARCHER project was initiated about 14 years ago to explore the use of emerging GPU technologies for fast Monte Carlo (MC) calculations. This paper presents the latest work to integrate the newly developed deep conventional neural network (dCNN) based MC denoising method with GPU-based MC multi-particle radiation transport simulation method to demonstrate a real-time dose computing capability for clinically realistic radiotherapy examples. The computing process involves GPU-based dose calculations that is followed by dCNN-based denoising. The dCNN-based dose denoiser is designed and employed to reduce the statistical uncertainty in dose distributions in patient anatomy defined by 3D computed tomography (CT) images. The training data include a range of dose distributions covering low-count/high-noise (DoseLCHN) and high-count/low-noise (DoseHCLN). The extremely large DoseLCHN and DoseHCLN dataset was generated from ARCHER. The DoseLCHN dataset is input into the trained model to output a predicted DoseHCLN dataset. For the evaluation, the DoseHCLN dataset produced by ARCHER is considered to be the ground truth. Experimental results show that the dose distributions generated from newly proposed method agreed consistently with the DoseHCLN produced from ARCHER. For hundreds of patient radiation treatment cases involving photons and protons, the average running time for one patient (GPU-based dose simulation followed by dCNN-based denoising) is about 200 ms. These preliminary results have demonstrated the feasibility of real-time Monte Carlo dose computing using an integrated dCNN-based denoising and GPU-based dose calculational approach. On-going studies involving more radiation types and clinical procedures are expected to facilitate the use of real-time MC dose planning and verification in the clinical workflow.

Evolutionary multi-objective algorithms have been applied to beam angle optimization (called BAO) for generating diverse trade-off radiotherapy treatment plans. However, their performance is not so effective due to the ignorance of using the specific clinical knowledge that can be obtain intuitively by clinical physicist. To address this issue, we suggest a pattern mining based evolutionary multi-objective algorithm called PM-EMA, in which two strategies for using the knowledge are proposed to accelerate the speed of population convergence. Firstly, to discover the potential beam angle distribution and discard the worse angles, the pattern mining strategy is used to detect the maximum and minimum sets of beam angles in non-dominated solutions of the population and utilize them to generate offspring to enhance the convergence. Moreover, to improve the quality of initial solutions, a tailored population initialization strategy is proposed by using the score of beam angles defined by this study. The experimental results on six clinical cancer cases demonstrate the superior performance of the proposed algorithm over six representative algorithms.

BackgroundIn pursuit of precise dose calculation and verification, the importance of beam modelling cannot be overstated, as it ensures an accurate distribution of particles incident upon the human body. The virtual source model, as one of the beam modelling methods, offers the advantage of not requiring detailed accelerator information. Although various virtual source models exist, manual adjustment to these models demands a substantial investment of time and computational resources. There has long been a desire to develop an efficient and automated approach for model commissioning.PurposeTo develop an automatic commissioning method for the virtual source model to customize the accelerator model for independent Monte Carlo dose verification.MethodsInitially, the accelerator model is established using the virtual source model and self-developed Jaw and MLC models. Then, a fully automated iteration process is employed to adjust the parameters of the virtual source model. Three types of objective functions are designed to represent differences from water tank measurements. Each objective function is paired with a specific parameter for adjustment, and their effectiveness is demonstrated through physical evidence. In each iteration, parameters with the highest objective function percentage are chosen for adjustment, and step length is determined based on current objective function values. Iteration is terminated when changes in any direction from the optimal solution no longer produce an improvement. Dose verification model for nine accelerators has been accomplished using this method. Additionally, under the same initial conditions, verification models for Versa HD accelerator (FF and FFF modes) are established using this method, Nelder-Mead Simplex optimization method, and the Bayesian optimization method to compare the efficiency and quality of these three iterative approaches.ResultsIterations for all nine accelerators are completed within 30 iterations. The relative dose differences in dose fall-off region compared to water tank measurements are all less than 2%, and the average gamma passing rates (3%/2 mm) for ArcCHECK measurements in QA plans are all higher than 97%. For Versa HD accelerator in FFF and FF modes, the proposed method achieves an average relative dose difference below 1% within 11 and 13 iterations, respectively. In contrast, the Simplex optimization reached 1% within 78 iterations in FFF mode. Furthermore, the Simplex optimization in FF mode and Bayesian optimization in both modes failed to achieve a 1% difference within 100 iterations.ConclusionsThe proposed iterative method achieves fast and automated commissioning of dose verification models, contributing to accurate and reliable clinical dose verification.

As a popular technique for intensity-modulated radiotherapy, direct aperture optimization (DAO) aims at generating treatment plans for cancer cases without the relaxation of optimization models. Conventional DAO methods are mainly based on mathematical programming, which can quickly generate a single plan but is inefficient in offering multiple candidate plans for clinical doctors. Recently, metaheuristics have been employed by DAO to offer many plans at a time; however, they are criticized for showing low efficiency in the evaluation and repair of iteratively generated offspring solutions. To provide an efficient DAO method, this work proposes a multi-objective evolutionary algorithm with customized variation operators. These operators can not only generate promising plans but also ensure their validity, and thus the search efficiency is improved by the acceleration of convergence and the elimination of repair operations. The experimental results demonstrate that the proposed DAO method is superior over existing heuristics and metaheuristics in terms of both effectiveness and efficiency.

AbstractThis paper presents the effort to extend a previously reported code ARCHER, a GPU‐based Monte Carlo (MC) code for coupled photon and electron transport, into protons including the consideration of magnetic fields. The proton transport is modeled using a Class‐II condensed‐history algorithm with continuous slowing‐down approximation. The model includes ionization, multiple scattering, energy straggling, elastic and inelastic nuclear interactions, as well as deflection due to the Lorentz force in magnetic fields. An additional direction change is added for protons at the end of each step in the presence of the magnetic field. Secondary charge particles, except for protons, are terminated depositing kinetic energies locally, whereas secondary neutral particles are ignored. Each proton is transported step by step until its energy drops to below 0.5 MeV or when the proton leaves the phantom. The code is implemented using the compute unified device architecture (CUDA) platform for optimized GPU thread‐level parallelism and efficiency. The code is validated by comparing it against TOPAS. Comparisons of dose distributions between our code and TOPAS for several exposure scenarios, ranging from single square beams in water to patient plan with magnetic fields, show good agreement. The 3D‐gamma pass rate with a 2 mm/2% criterion in the region with dose greater than 10% of the maximum dose is computed to be over 99% for all tested cases. Using a single NVIDIA TITAN V GPU card, the computational time of ARCHER is found to range from 0.82 to 4.54 seconds for 1 × 107 proton histories. Compared to a few hours running on TOPAS, this speed improvement is significant. This work presents, for the first time, the performance of a GPU‐based MC code to simulate proton transportation magnetic fields, demonstrating the feasibility of accurate and efficient dose calculations in potential magnetic resonance imaging (MRI)‐guided proton therapy.

Purpose To apply an independent GPU-accelerated Monte Carlo (MC) dose verification for CyberKnife M6 with Iris collimator and evaluate the dose calculation accuracy of RayTracing (TPS-RT) algorithm and Monte Carlo (TPS-MC) algorithm in the Precision treatment planning system (TPS). Methods GPU-accelerated MC algorithm (ArcherQA-CK) was integrated into a commercial dose verification system, ArcherQA, to implement the patient-specific quality assurance in the CyberKnife M6 system. 30 clinical cases (10 cases in head, and 10 cases in chest, and 10 cases in abdomen) were collected in this study. For each case, three different dose calculation methods (TPS-MC, TPS-RT and ArcherQA-CK) were implemented based on the same treatment plan and compared with each other. For evaluation, the 3D global gamma analysis and dose parameters of the target volume and organs at risk (OARs) were analyzed comparatively. Results For gamma pass rates at the criterion of 2%/2 mm, the results were over 98.0% for TPS-MC vs.TPS-RT, TPS-MC vs. ArcherQA-CK and TPS-RT vs. ArcherQA-CK in head cases, 84.9% for TPS-MC vs.TPS-RT, 98.0% for TPS-MC vs. ArcherQA-CK and 83.3% for TPS-RT vs. ArcherQA-CK in chest cases, 98.2% for TPS-MC vs.TPS-RT, 99.4% for TPS-MC vs. ArcherQA-CK and 94.5% for TPS-RT vs. ArcherQA-CK in abdomen cases. For dose parameters of planning target volume (PTV) in chest cases, the deviations of TPS-RT vs. TPS-MC and ArcherQA-CK vs. TPS-MC had significant difference (P < 0.01), and the deviations of TPS-RT vs. TPS-MC and TPS-RT vs. ArcherQA-CK were similar (P > 0.05). ArcherQA-CK had less calculation time compared with TPS-MC (1.66 min vs. 65.11 min). Conclusions Our proposed MC dose engine (ArcherQA-CK) has a high degree of consistency with the Precision TPS-MC algorithm, which can quickly identify the calculation errors of TPS-RT algorithm for some chest cases. ArcherQA-CK can provide accurate patient-specific quality assurance in clinical practice.

Intensity-modulate proton therapy is one of the most advanced cancer treatment techniques due to the Bragg peak characteristics of proton radiation. The personalized demand of different patients requires treatment optimization methods to quickly provide diverse treatment plans to select the best plan for a patient. However, most existing treatment optimization methods are transformed the multi-objective optimization problem into a single optimization problem. Moreover, the radiation physicists may adjust the objective weights repeatedly to produce a set of high-quality treatment plans. To address this problem, this paper proposes an adaptive conjugate gradient accelerated evolutionary algorithm (ACG-EA) to generate a set of diverse high-quality treatment plans simultaneously. The conjugate gradient method is employed as a directional mutation operator to accelerate the search process in the hybrid mutation operation. In addition, the weight parameters of the conjugate gradient are automatically updated based on the diversity and convergence of the current population. Compared with five representative multi-objective evolutionary algorithms, the experimental results have shown the competitive performance of the proposed ACG-EA on the hypervolume and dose-volume histogram indicators in six clinical cancer cases.

Objective: Adaptive planning to accommodate anatomic changes during treatment often requires repeated segmentation. In this study, prior patient-specific data was integrateda into a registration-guided multi-channel multi-path (Rg-MCMP) segmentation framework to improve the accuracy of repeated clinical target volume (CTV) segmentation. Methods: This study was based on CT image datasets for a total of 90 cervical cancer patients who received two courses of radiotherapy. A total of 15 patients were selected randomly as the test set. In the Rg-MCMP segmentation framework, the first-course CT images (CT1) were registered to second-course CT images (CT2) to yield aligned CT images (aCT1), and the CTV in the first course (CTV1) was propagated to yield aligned CTV contours (aCTV1). Then, aCT1, aCTV1, and CT2 were combined as the inputs for 3D U-Net consisting of a channel-based multi-path feature extraction network. The performance of the Rg-MCMP segmentation framework was evaluated and compared with the single-channel single-path model (SCSP), the standalone registration methods, and the registration-guided multi-channel single-path (Rg-MCSP) model. The Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and average surface distance (ASD) were used as the metrics. Results: The average DSC of CTV for the deformable image DIR-MCMP model was found to be 0.892, greater than that of the standalone DIR (0.856), SCSP (0.837), and DIR-MCSP (0.877), which were improvements of 4.2%, 6.6%, and 1.7%, respectively. Similarly, the rigid body DIR-MCMP model yielded an average DSC of 0.875, which exceeded standalone RB (0.787), SCSP (0.837), and registration-guided multi-channel single-path (0.848), which were improvements of 11.2%, 4.5%, and 3.2%, respectively. These improvements in DSC were statistically significant (p < 0.05). Conclusion: The proposed Rg-MCMP framework achieved excellent accuracy in CTV segmentation as part of the adaptive radiotherapy workflow.