4DCT Datasets Research Articles

Technical note: Performance evaluation of volumetric imaging based on motion modeling by principal component analysis.

To quantitatively evaluate the achievable performance of volumetric imaging based on lung motion modeling by principal component analysis (PCA). In volumetric imaging based on PCA, internal deformation was represented as a linear combination of the eigenvectors derived by PCA of the deformation vector fields evaluated from patient-specific four-dimensional-computed tomography (4DCT) datasets. The volumetric image was synthesized by warping the reference CT image with a deformation vector field which was evaluated using optimal principal component coefficients (PCs). Larger PCs were hypothesized to reproduce deformations larger than those included in the original 4DCT dataset. To evaluate the reproducibility of PCA-reconstructed volumetric images synthesized to be close to the ground truth as possible, mean absolute error (MAE), structure similarity index measure (SSIM) and discrepancy of diaphragm position were evaluated using 22 4DCT datasets of nine patients. Mean MAE and SSIM values for the PCA-reconstructed volumetric images were approximately 80HU and 0.88, respectively, regardless of the respiratory phase. In most test cases including the data of which motion range was exceeding that of the modeling data, the positional error of diaphragm was less than 5 mm. The results suggested that large deformations not included in the modeling 4DCT dataset could be reproduced. Furthermore, since the first PC correlated with the displacement of the diaphragm position, the first eigenvector became the dominant factor representing the respiration-associated deformations. However, other PCs did not necessarily change with the same trend as the first PC, and no correlation was observed between the coefficients. Hence, randomly allocating or sampling these PCs in expanded ranges may be applicable to reasonably generate an augmented dataset with various deformations. Reasonable accuracy of image synthesis comparable to those in the previous research were shown by using clinical data. These results indicate the potential of PCA-based volumetric imaging for clinical applications.

Investigation of four-dimensional (4D) Monte Carlo dose calculation in real-time tumor tracking stereotatic body radiotherapy for lung cancers.

To investigate the dosimetric variations and radiobiological impacts as a consequence of delivering treatment plans of 3D nature in 4D manner based on the 4D Monte Carlo treatment planning framework implemented on Cyberknife. Dose distributions were optimized on reference 3D images at end of exhale phase of a 4DCT dataset for twenty-five lung cancer patients treated with 60 Gy / 3Fx or 48 Gy / 4Fx. Deformable image registrations (DIR) between individual 3DCT images to the reference 3DCT image in the 4DCT study were performed to interpolate doses calculated on multiple anatomical geometries back on to the reference geometry to compose a 4D dose distribution that included the tracking beam motion and organ deformation. The 3D and 4D dose distributions that were initially calculated with the equivalent path-length (EPL) algorithm (3DEPL dose and 4DEPL dose) were recalculated with the Monte Carlo algorithm (3DMC dose and 4DMC dose). Dosimetric variations of V60Gy / 48Gy and D99 of GTV, mean doses to the lung and the heart and maximum dose (D1 ) of the spinal cord as a consequence of tracking beam motion in deforming anatomy, dose calculation algorithm, and both were quantified by the relative change from 4DMC to 3DMC doses, from 4DMC to 4DEPL doses, and from 4DMC to 3DEPL doses, respectively. Comparing 4DMC to 3DEPL plans, V60Gy / 48Gy and D99 of GTV decreased considerably by 13 ± 22% (mean ± 1SD) and 9.2 ± 5.5 Gy but changes of normal tissue doses were not more than 0.5 Gy on average. The generalized equivalent uniform dose (gEUD) and tumor control probability (TCP) were reduced by 14.3 ± 8.8 Gy and 7.5 ± 5.2%, and normal tissue complication probability (NTCP) for myelopathy and pericarditis were close to zero and NTCP for radiation pneumonitis was reduced by 2.5 ± 4.1%. Comparing 4DMC to 4DEPL plans found decreased V60Gy / 48Gy and D99 by 12.3 ± 21.6% and 7.3 ± 5.3 Gy, the normal tissues doses by 0.5 Gy on average, gEUD and TCP by 13.0 ± 8.6 Gy and 7.1 ± 5.1%. Comparing 4DMC to 3DMC doses, V60Gy / 48Gy and D99 of GTV was reduced by 5.2 ± 8.8 %and 2.6 ± 3.3 Gy, and normal tissues hardly changed from 4DMC to 3DMC doses. The corresponding decreases of gEUD and TCP were 2.8 ± 4.0 Gy and 1.6 ± 2.4%. The large discrepancy between original 3DEPL plan and benchmarking 4DMC plan is predominately due to dose calculation algorithms as the tracking beam motion and organ deformation hardly influenced doses of normal tissues and moderately decreased V60Gy / 48Gy and D99 of GTV. It is worth to make a thoughtful weight of the benefits of full 4D MC dose calculation and consider 3D MC dose calculation as a compromise of 4D MC dose calculation considering the multifold computation time. This article is protected by copyright. All rights reserved.

PLOSL: Population learning followed by one shot learning pulmonary image registration using tissue volume preserving and vesselness constraints.

This paper presents the Population Learning followed by One Shot Learning (PLOSL) pulmonary image registration method. PLOSL is a fast unsupervised learning-based framework for 3D-CT pulmonary image registration algorithm based on combining population learning (PL) and one-shot learning (OSL). The PLOSL image registration has the advantages of the PL and OSL approaches while reducing their respective drawbacks. The advantages of PLOSL include improved performance over PL, substantially reducing OSL training time and reducing the likelihood of OSL getting stuck in local minima. PLOSL pulmonary image registration uses tissue volume preserving and vesselness constraints for registration of inspiration-to-expiration and expiration-to-inspiration pulmonary CT images. A coarse-to-fine convolution encoder-decoder CNN architecture is used to register large and small shape features. During training, the sum of squared tissue volume difference (SSTVD) compensates for intensity differences between inspiration and expiration computed tomography (CT) images and the sum of squared vesselness measure difference (SSVMD) helps match the lung vessel tree. Results show that the PLOSL (SSTVD+SSVMD) algorithm achieved subvoxel landmark error while preserving pulmonary topology on the SPIROMICS data set, the public DIR-LAB COPDGene and 4DCT data sets.

Validation of an established deep learning auto-segmentation tool for cardiac substructures in 4D radiotherapy planning scans.

Emerging data suggest that dose-sparing several key cardiac regions is prognostically beneficial in lung cancer radiotherapy. The cardiac substructures are challenging to contour due to their complex geometry, poor soft tissue definition on computed tomography (CT) and cardiorespiratory motion artefact. A neural network was previously trained to generate the cardiac substructures using three-dimensional radiotherapy planning CT scans (3D-CT). In this study, the performance of that tool on the average intensity projection from four-dimensional (4D) CT scans (4D-AVE), now commonly used in lung radiotherapy, was evaluated. The 4D-AVE of n=20 patients completing radiotherapy for lung cancer 2015-2020 underwent manual and automated cardiac substructure segmentation. Manual and automated substructures were compared geometrically and dosimetrically. Two senior clinicians also qualitatively assessed the auto-segmentation tool's output. Geometric comparison of the automated and manual segmentations exhibited high levels of similarity across parameters, including volume difference (11.8% overall) and Dice similarity coefficient (0.85 overall), and were consistent with 3D-CT performance. Differences in mean (median 0.2Gy, range -1.6-0.3Gy) and maximum (median 0.4Gy, range -2.2-0.9Gy) doses to substructures were generally small. Nearly all structures (99.5%) were deemed to be appropriate for clinical use without further editing. Cardiac substructure auto-segmentation using a deep learning-based tool trained on a 3D-CT dataset was feasible on the 4D-AVE scan, meaning this tool is suitable for use on 4D-CT radiotherapy planning scans. Application of this tool would increase the practicality of routine clinical cardiac substructure delineation, and enable further cardiac radiation effects research.

Evaluation of target volume and dose accuracy in intrafractional cases of lung cancer based on 4D-CT and 4D-CBCT images using an in-house dynamic thorax phantom.

BackgroundThis study aimed to evaluate the target volume and dose accuracy in intrafraction cases using 4-dimensional imaging modalities and an in-house dynamic thorax phantom. Intrafraction motion can create errors in the definition of target volumes, which can significantly affect the accuracy of radiation delivery. Motion management using 4-dimensional modalities is required to reduce the risk.Materials and methodsTwo variations in both breathing amplitude and target size were applied in this study. From these variations, internal target volume (ITVs) contoured in 10 phases of 4D-CT (ITV10), average intensity projection (AIP), and mid-ventilation (Mid-V) images were reconstructed from all 4D-CT datasets as reference images. Free-breathing (FB), augmentation free-breathing (Aug-FB), and static images were also acquired using the 3D-CT protocol for comparisons. In dose evaluations, the 4D-CBCT modality was applied before irradiation to obtain position correction. Then, the dose was evaluated with Gafchromic film EBT3.ResultsThe ITV10, AIP, and Mid-V provide GTVs that match the static GTV. The AIP and Mid-V reference images allowed reductions in ITVs and PTVs without reducing the range of target movement areas compared to FB and Aug-FB images with varying percentages in the range of 29.17% to 48.70%. In the dose evaluation, the largest discrepancies between the measured and planned doses were 10.39% for the FB images and 9.21% for the Aug-FB images.ConclusionThe 4D-CT modality can enable accurate definition of the target volume and reduce the PTV. Furthermore, 4D-CBCT provides localization images during registration to facilitate position correction and accurate dose delivery.

Dosimetric Planning Comparison for Left-Sided Breast Cancer Radiotherapy: The Clinical Feasibility of Four-Dimensional-Computed Tomography-Based Treatment Planning Optimization.

Background: Adjuvant whole-breast radiotherapy (RT) is a significant part of the standard of care treatment after breast cancer (BC) conserving surgery. Modern techniques including intensity-modulated radiation therapy (IMRT) and volumetric-modulated arc therapy (VMAT) have constituted to better target coverage and critical organs sparing. However, BC survivors are at risk of developing radiation-induced cardiac toxicity. Hence, deep-inspiration breath-hold (DIBH) techniques have been implemented at many centers to further reduce cardiac exposure but require compliance. 4D-CT robust optimization can account for heart intrafractional motion per breathing phase. The optimization has been explored in cardiac sparing of breast IMRT compared to DIBH in a small sample size but has not been evaluated in substructures sparing, nor in VMAT. To provide patients who are not compliant to breath-hold with an optimal treatment approach, various heart sparing techniques need to be evaluated for statistical significance and clinical feasibility.Aim: This retrospective study aimed to provide an extensive dosimetric heart sparing comparison of free-breathing, 4D-CT-based treatment planning, including robust optimization with DIBH-based treatment planning. Combinations of forward and inverse IMRT and VMAT are also considered.Methods: Fifteen early stage left-sided BC standard treatment plans were selected. Breast, lung, left anterior descending artery (LAD), left ventricle (LV), and the whole heart were contoured on each 4D-CT phase and DIBH CT dataset. Each treatment plan was optimized using forward/inverse IMRT and VMAT on the following CT datasets: DIBH, average 4D-CT, and the complete 4D-CT dataset needed for robust optimization. Dose-volume histograms were used to compare V5GyHeart, mean heart dose, mean and max LAD dose, mean LV dose, and V50%Lung.Results: All RT techniques assessed including 4D robust optimization were clinically feasible. Statistically significant differences in mean heart, LAD and LV dose, max LAD dose, and V5GyHeart (p < 0.01) but no difference in V50%Lung (p = 0.29) were found between different techniques. IMRT DIBH achieved the optimal cardiac and substructure sparing among treatment plans. 4D robust IMRT had significantly greater mean heart and LV dose than DIBH IMRT (p ≤ 0.01), except LAD dose. Among free-breathing methods, no difference in all cardiac and substructure dose parameters was observed (p > 0.2) in comparing forward and inverse IMRT with average 4D-CT, inverse average 4D-CT, and 4D robust with IMRT, and between average 4D-CT VMAT and 4D robust VMAT. Only V5GyHeart and mean LV dose were significantly greater in 4D robust VMAT (p < 0.01) compared to DIBH VMAT. Mean heart and LV doses were significantly reduced (p < 0.01) in DIBH IMRT compared to DIBH VMAT. Moreover, mean heart and LV dose, V5GyHeart were significantly reduced in inverse IMRT average 4D-CT compared to average 4D-CT VMAT (p < 0.02) and in 4D robust IMRT compared to 4D robust VMAT (p < 0.04).Conclusion: This study demonstrated the clinical feasibility of 4D robust optimization in limiting the cardiac and substructures dose during free-breathing RT with both IMRT/VMAT for patients who are not compliant with breath-hold RT. However, this study also presents that 4D robust optimization can reduce LAD dose but not fully outperform DIBH or conventional 4D-CT-based planning with IMRT/VMAT in heart sparing in treating early staged left-sided BC patients.

Patient-specific deep learning model to enhance 4D-CBCT image for radiomics analysis

Objective. 4D-CBCT provides phase-resolved images valuable for radiomics analysis for outcome prediction throughout treatment courses. However, 4D-CBCT suffers from streak artifacts caused by under-sampling, which severely degrades the accuracy of radiomic features. Previously we developed group-patient-trained deep learning methods to enhance the 4D-CBCT quality for radiomics analysis, which was not optimized for individual patients. In this study, a patient-specific model was developed to further improve the accuracy of 4D-CBCT based radiomics analysis for individual patients. Approach. This patient-specific model was trained with intra-patient data. Specifically, patient planning 4D-CT was augmented through image translation, rotation, and deformation to generate 305 CT volumes from 10 volumes to simulate possible patient positions during the onboard image acquisition. 72 projections were simulated from 4D-CT for each phase and were used to reconstruct 4D-CBCT using FDK back-projection algorithm. The patient-specific model was trained using these 305 paired sets of patient-specific 4D-CT and 4D-CBCT data to enhance the 4D-CBCT image to match with 4D-CT images as ground truth. For model testing, 4D-CBCT were simulated from a separate set of 4D-CT scan images acquired from the same patient and were then enhanced by this patient-specific model. Radiomics features were then extracted from the testing 4D-CT, 4D-CBCT, and enhanced 4D-CBCT image sets for comparison. The patient-specific model was tested using 4 lung-SBRT patients’ data and compared with the performance of the group-based model. The impact of model dimensionality, region of interest (ROI) selection, and loss function on the model accuracy was also investigated. Main results. Compared with a group-based model, the patient-specific training model further improved the accuracy of radiomic features, especially for features with large errors in the group-based model. For example, the 3D whole-body and ROI loss-based patient-specific model reduces the errors of the first-order median feature by 83.67%, the wavelet LLL feature maximum by 91.98%, and the wavelet HLL skewness feature by 15.0% on average for the four patients tested. In addition, the patient-specific models with different dimensionality (2D versus 3D) or loss functions (L1 versus L1 + VGG + GAN) achieved comparable results for improving the radiomics accuracy. Using whole-body or whole-body+ROI L1 loss for the model achieved better results than using the ROI L1 loss alone as the loss function. Significance. This study demonstrated that the patient-specific model is more effective than the group-based model on improving the accuracy of the 4D-CBCT radiomic features analysis, which could potentially improve the precision for outcome prediction in radiotherapy.

Quantitative analysis of dose-averaged linear energy transfer (LETd ) robustness in pencil beam scanning proton lung plans.

The primary objective of our study was to perform a quantitative robustness analysis of the dose-averaged linear energy transfer (LETd ) and related RBE-weighted dose in robustly optimized (in terms of the range and set up uncertainties) pencil beam scanning (PBS) proton lung cancer plans. In this study, we utilized the 4DCT dataset of six anonymized lung patients. PBS lung plans were generated using a robust optimization technique (range uncertainty: ±3.5% and setup errors: ±5mm) on the CTV for a total dose of 5000 cGy (RBE) in five fractions using the RBE of 1.1. For each patient, the LETd distributions were calculated for the nominal plan and three groups. Group 1: two plan robustness scenarios for range uncertainties of ±3.5%; Group 2: twelve plan robustness scenarios (range uncertainty (±3.5%) in conjunction with setup errors (±5mm)); and Group 3: ten different breathing phases of the 4DCT dataset. The RBE-weighted dose to the OARs was evaluated for all robustness scenarios and breathing phases. The variation (∆) in the mean LETd and mean RBE-weighted dose from each group was recorded. The mean LETd in the CTV of nominal PBS lung plans among six patients ranged from 2.2 to 2.6keV/µm. On average, for the combined range and setup uncertainties, the ∆ in the mean LETd among 12 scenarios of all six patients was 0.6keV/µm, which is slightly higher than when only the range uncertainties were considered (0.4keV/µm). The ∆ in the mean LETd in a patient was ≤1.7keV/µm in the heart and ≤1.2keV/µm in the esophagus and total lung. The ∆ in the mean RBE-weighted dose in a patient was up to 79 cGy for the total lung, 165 cGy for the heart, and 258 cGy for the esophagus. For ten breathing phases, the ∆ in the mean LETd in a patient was ≤0.3keV/µm in the CTV, ≤0.5keV/µm in the heart, ≤0.4keV/µm in the esophagus, and ≤0.7keV/µm in the total lung. The addition of setup errors to the range uncertainties resulted in slightly less homogeneous LETd distributions. The variations in the mean LETd among the ten breathing phases were slightly larger in the total lung than in the heart and esophagus. The combination of setup and range uncertainties had a greater impact than the effect of breathing phases on the variations in the mean RBE-weighted dose to the OARs. Overall, the LETd distributions in the CTV were less sensitive than those in the OARs to setup errors, range uncertainties, and breathing phases for robustly optimized (in terms of range and setup uncertainities) PBS proton lung cancer plans.

Proceedings to the 7th Annual Conference of the Particle Therapy Cooperative Group North America (PTCOG-NA)

Proceedings to the 7th Annual Conference of the Particle Therapy Cooperative Group North America (PTCOG-NA)

Impact of Different Synchrotron Flattop Operation Modes on 4D Dosimetric Uncertainties for Scanned Carbon-Ion Beam Delivery.

PurposeThe characteristic of pulsed beam delivery for synchrotron-based carbon-ion radiotherapy has led to the emergence of many scanning scenarios in order to improve the treatment efficiency and accuracy of moving target volume. Here, we aim to evaluate a novel breathing guidance motion mitigation performance under different synchrotron flattop operation modes in carbon-ion radiotherapy.MethodsWith the use of twelve 4DCT datasets of lung cancer patients who had been treated with respiratory-gated carbon-ion pencil beam therapy, range-adapted internal target volume (raITV) plans were optimized. Under the fixed flattop with single-energy and extended flattop with multi-energy synchrotron operation modes, the 4D treatments with breathing guidance and free breathing-based gated phase-controlled rescanning (PCR) beam delivery were simulated. Dose metrics (D95 and D5–D95 in clinical target volume (CTV)) and treatment time of the resulting 4D plans were compared.ResultsThe two synchrotron operation modes provided different scanning dynamics. For the free breathing-based PCR method delivered in the extended flattop operation mode, the averaged CTV-D95 values were 90.4% ± 3.7%, 95.4% ± 1.7%, 96.9% ± 1.5%, 97.2% ± 1.5%, and 97.3% ± 1.5% for the 1-scanning, 2-PCR, 4-PCR, 6-PCR, and 8-PCR, respectively. For the breathing guidance-based PCR method delivered in the extended flattop mode, these values were 89.1% ± 4.0%, 97.0% ± 1.4%, 98.2% ± 0.7%, 98.6% ± 0.7%, and 98.9% ± 0.7%, respectively. However, CTV-D95 significantly increased to 98.5% ± 1.0% even with just 1-scanning breathing guidance-based fixed flattop operation mode (p < 0.01). Moreover, there was no significant difference in treatment time among the three technical combinations (p > 0.15).ConclusionsThe combination of the breathing guidance and PCR methods should be an alternative way for motion mitigation for the fixed flattop synchrotron operation mode. The target dose coverage and homogeneity could be further improved by the combination of the breathing guidance and PCR methods than the traditional PCR-only technology for the extended flattop synchrotron operation mode.

Adaptive Hierarchical Dual Consistency for Semi-Supervised Left Atrium Segmentation on Cross-Domain Data.

Semi-supervised learning provides great significance in left atrium (LA) segmentation model learning with insufficient labelled data. Generalising semi-supervised learning to cross-domain data is of high importance to further improve model robustness. However, the widely existing distribution difference and sample mismatch between different data domains hinder the generalisation of semi-supervised learning. In this study, we alleviate these problems by proposing an Adaptive Hierarchical Dual Consistency (AHDC) for the semi-supervised LA segmentation on cross-domain data. The AHDC mainly consists of a Bidirectional Adversarial Inference module (BAI) and a Hierarchical Dual Consistency learning module (HDC). The BAI overcomes the difference of distributions and the sample mismatch between two different domains. It mainly learns two mapping networks adversarially to obtain two matched domains through mutual adaptation. The HDC investigates a hierarchical dual learning paradigm for cross-domain semi-supervised segmentation based on the obtained matched domains. It mainly builds two dual-modelling networks for mining the complementary information in both intra-domain and inter-domain. For the intra-domain learning, a consistency constraint is applied to the dual-modelling targets to exploit the complementary modelling information. For the inter-domain learning, a consistency constraint is applied to the LAs modelled by two dual-modelling networks to exploit the complementary knowledge among different data domains. We demonstrated the performance of our proposed AHDC on four 3D late gadolinium enhancement cardiac MR (LGE-CMR) datasets from different centres and a 3D CT dataset. Compared to other state-of-the-art methods, our proposed AHDC achieved higher segmentation accuracy, which indicated its capability in the cross-domain semi-supervised LA segmentation.

Lung-CRNet: A convolutional recurrent neural network for lung 4DCT image registration.

Deformable image registration (DIR) of lung four-dimensional computed tomography (4DCT) plays a vital role in a wide range of clinical applications. Most of the existing deep learning-based lung 4DCT DIR methods focus on pairwise registration which aims to register two images with large deformation. However, the temporal continuities of deformation fields between phases are ignored. This paper proposes a fast and accurate deep learning-based lung 4DCT DIR approach that leverages the temporal component of 4DCT images. We present Lung-CRNet, an end-to-end convolutional recurrent registration neural network for lung 4DCT images and reformulate 4DCT DIR as a spatiotemporal sequence predicting problem in which the input is a sequence of three-dimensional computed tomography images from the inspiratory phase to the expiratory phase in a respiratory cycle. The first phase in the sequence is selected as the only reference image and the rest as moving images. Multiple convolutional gated recurrent units (ConvGRUs) are stacked to capture the temporal clues between images. The proposed network is trained in an unsupervised way using a spatial transformer layer. During inference, Lung-CRNet is able to yield the respective displacement field for each reference-moving image pair in the input sequence. We have trained the proposed network using a publicly available lung 4DCT dataset and evaluated performance on the widely used the DIR-Lab dataset. The mean and standard deviation of target registration error are 1.56 1.05 mm on the DIR-Lab dataset. The computation time for each forward prediction is less than 1 s on average. The proposed Lung-CRNet is comparable to the existing state-of-the-art deep learning-based 4DCT DIR methods in both accuracy and speed. Additionally, the architecture of Lung-CRNet can be generalized to suit other groupwise registration tasks which align multiple imagessimultaneously.

Few-shot learning for deformable image registration in 4DCT images.

To develop a rapid and accurate 4D deformable image registration (DIR) approach for online adaptive radiotherapy. We propose a deep learning (DL)-based few-shot registration network (FR-Net) to generate deformation vector fields from each respiratory phase to an implicit reference image, thereby mitigating the bias introduced by the selection of reference images. The proposed FR-Net is pretrained with limited unlabeled 4D data and further optimized by maximizing the intensity similarity of one specific four-dimensional computed tomography (4DCT) scan. Because of the learning ability of DL models, the few-shot learning strategy facilitates the generalization of the model to other 4D data sets and the acceleration of the optimization process. The proposed FR-Net is evaluated for 4D groupwise and 3D pairwise registration on thoracic 4DCT data sets DIR-Lab and POPI. FR-Net displays an averaged target registration error of 1.48 mm and 1.16 mm between the maximum inhalation and exhalation phases in the 4DCT of DIR-Lab and POPI, respectively, with approximately 2 min required to optimize one 4DCT. Overall, FR-Net outperforms state-of-the-art methods in terms of registration accuracy and exhibits a low computational time. We develop a few-shot groupwise DIR algorithm for 4DCT images. The promising registration performance and computational efficiency demonstrate the prospective applications of this approach in registration tasks for online adaptive radiotherapy. This work exploits DL models to solve the optimization problem in registering 4DCT scans while combining groupwise registration and few-shot learning strategy to solve the problem of consuming computational time and inferior registration accuracy.

Modified VMAT Plans for Locally Advanced Centrally Located Non-Small Cell Lung Cancer (NSCLC).

Objectives: This study aimed to find the optimal radiotherapy VMAT plans, that achieved high conformity and homogeneity to the planned target volume (PTV), and minimize the dose to nearby organs at risk including the non-PTV lung, heart and oesophagus for patients with centrally located non-small Cell Lung Cancer. Methods: A total of 18 patients who were treated for stage III centrally located non-small Cell Lung Cancer were selected retrospectively for this study. Identical CT datasets, 4D CT and structure dataset were used for radiotherapy planning based on single-planar VMAT (SP-VMAT), dual-planar VMAT (DP-VMAT) and Hybrid VMAT (H-VMAT). For SP-VMAT, one full arc and two half arcs were created on single-plane with couch at 0°. For DP-VMAT, one full arc was created with couch at 0°, and two half arcs with couch rotation of 330° or 30°. For H-VMAT, anterior-posterior opposing fixed beam and two half arcs were planned at couch at 0°. Dose constraints were adhered to the RTOG0617. Dose volumetric parameters were collected for statistical analysis. Results: There were no significant differences for the PTV, HI, CI between the SP-VMAT, DP-VMAT and H-VMAT. For the non-PTV lungs, Dmean, V20, V10, V5, D1500 and D1000 were significantly lower (2.05 Gy, 6.47%, 15.89%, 11.66% 4.17 Gy and 5.47 Gy respectively) in H-VMAT than that of SP-VMAT (all p < 0.001). For the oesophagus, Dmax, Dmean, V30 and V18.8 of H-VMAT were 0.08 Gy, 1.73 Gy, 5.54% and 7.17% lower than that of the SP-VMAT plan. For the heart, Dmean, V34, V28, V20 and V10 of DP-VMAT were lower than that of SP-VMAT by 1.45 Gy, 0.65%, 1.74%, 4.8% and 7.11% respectively. Conclusion: The proposed H-VMAT showed more favourable plan quality than the SP-VMAT for centrally located stage III NSCLC, in particular for non-PTV lungs and the oesophagus. It will benefit patients, especially those who planned for immunotherapy (Durvalumab) after standard chemo-irradiation. The proposed DP-VMAT plan showed significant dose reduction to the heart when compared to the H-VMAT plan.

Impact of respiratory motion in dosimetric and clinical advantages for adjuvant left-sided breast radiotherapy.

We investigated the organ-sparing effect of the deep inspiration breath hold (DIBH) technique among different levels of lung expansion for left-side breast radiotherapy. This retrospective study enrolled 30 patients who received adjuvant left breast radiotherapy after breast-conserving surgery (BCS). Simulation scans of both DIBH and deep breathing four-dimensional computed tomography (4DCT) were acquired, and three treatment plans were generated for each patient. One plan was based on the DIBH images, and the other two plans were based on the mid-lung expansion (ME) and initial lung expansion (IE) phases retrieved from 4DCT data sets. Dosimetric comparisons and normal tissue complication probability (NTCP) models were conducted. We used image registration for displacement analysis and sought potential factors related to the dose benefit of DIBH. The DIBH plans resulted significantly lower doses to the heart, left ventricle (LV) and left anterior descending coronary artery (LAD), including the high- to low-dose areas, followed by the ME plans and IE plans (p < 0.05). DIBH reduced the risk of long-term cardiac mortality by 40% and radiation pneumonitis of the left lung by 37.96% compared with the IE plans (p < 0.001). The reduction in the mean dose to the heart and LV significantly correlated with anterior displacement of the left lung. The DIBH technique is a feasible tool to provide dosimetric and clinical advantages for adjuvant left-sided breast radiotherapy. Breathing pattern and the level of lung expansion seem to play an important role.

Lung tumor segmentation in 4D CT images using motion convolutional neural networks.

Manual delineation on all breathing phases of lung cancer 4D CT image datasets can be challenging, exhaustive, and prone to subjective errors because of both the large number of images in the datasets and variations in the spatial location of tumors secondary to respiratory motion. The purpose of this work is to present a new deep learning-based framework for fast and accurate segmentation of lung tumors on 4D CT image sets. The proposed DL framework leverages motion region convolutional neural network (R-CNN). Through integration of global and local motion estimation network architectures, the network can learn both major and minor changes caused by tumor motion. Our network design first extracts tumor motion information by feeding 4D CT images with consecutive phases into an integrated backbone network architecture, locating volume-of-interest (VOIs) via a regional proposal network and removing irrelevant information via a regional convolutional neural network. Extracted motion information is then advanced into the subsequent global and local motion head network architecture to predict corresponding deformation vector fields (DVFs) and further adjust tumor VOIs. Binary masks of tumors are then segmented within adjusted VOIs via a mask head. A self-attention strategy is incorporated in the mask head network to remove any noisy features that might impact segmentation performance. We performed two sets of experiments. In the first experiment, a five-fold cross-validation on 20 4D CT datasets, each consisting of 10 breathing phases (i.e., 200 3D image volumes in total). The network performance was also evaluated on an additional unseen 200 3D images volumes from 20 hold-out 4D CT datasets. In the second experiment, we trained another model with 40 patients' 4D CT datasets from experiment 1 and evaluated on additional unseen nine patients' 4D CT datasets. The Dice similarity coefficient (DSC), center of mass distance (CMD), 95th percentile Hausdorff distance (HD95 ), mean surface distance (MSD), and volume difference (VD) between the manual and segmented tumor contour were computed to evaluate tumor detection and segmentation accuracy. The performance of our method was quantitatively evaluated against four different methods (VoxelMorph, U-Net, network without global and local networks, and network without attention gate strategy) across all evaluation metrics through a paired t-test. The proposed fully automated DL method yielded good overall agreement with the ground truth for contoured tumor volume and segmentation accuracy. Our model yielded significantly better values of evaluation metrics (p<0.05) than all four competing methods in both experiments. On hold-out datasets of experiment 1 and 2, our method yielded DSC of 0.86 and 0.90 compared to 0.82 and 0.87, 0.75 and 0.83, 081 and 0.89, and 0.81 and 0.89 yielded by VoxelMorph, U-Net, network without global and local networks, and networks without attention gate strategy. Tumor VD between ground truth and our method was the smallest with the value of 0.50 compared to 0.99, 1.01, 0.92, and 0.93 for between ground truth and VoxelMorph, U-Net, network without global and local networks, and networks without attention gate strategy, respectively. Our proposed DL framework of tumor segmentation on lung cancer 4D CT datasets demonstrates a significant promise for fully automated delineation. The promising results of this work provide impetus for its integration into the 4D CT treatment planning workflow to improve the accuracy and efficiency of lung radiotherapy.

Real-time CT image generation based on voxel-by-voxel modeling of internal deformation by utilizing the displacement of fiducial markers.

To show the feasibility of real-time CT image generation technique utilizing internal fiducial markers that facilitate the evaluation of internal deformation. In the proposed method, a linear regression model that can derive internal deformation from the displacement of fiducial markers is built for each voxel in the training process before the treatment session. Marker displacement and internal deformation are derived from the four-dimensional computed tomography (4DCT) dataset. In the treatment session, the three-dimensional deformation vector field is derived according to the marker displacement, which is monitored by the real-time imaging system. The whole CT image can be synthesized by deforming the reference CT image with a deformation vector field in real-time. To show the feasibility of the technique, image synthesis accuracy and tumor localization accuracy were evaluated using the dataset generated by extended NURBS-Based Cardiac-Torso (XCAT) phantom and clinical 4DCT datasets from six patients, containing 10 CT datasets each. In the validation with XCAT phantom, motion range of the tumor in training data and validation data were about 10 and 15mm, respectively, so as to simulate motion variation between 4DCT acquisition and treatment session. In the validation with patient 4DCT dataset, eight CT datasets from the 4DCT dataset were used in the training process. Two excluded inhale CT datasets can be regarded as the datasets with large deformations more than training dataset. CT images were generated for each respiratory phase using the corresponding marker displacement. Root mean squared error (RMSE), normalized RMSE (NRMSE), and structural similarity index measure (SSIM) between the original CT images and the synthesized CT images were evaluated as the quantitative indices of the accuracy of image synthesis. The accuracy of tumor localization was also evaluated. In the validation with XCAT phantom, the mean NRMSE, SSIM, and three-dimensional tumor localization error were 7.5±1.1%, 0.95±0.02, and 0.4±0.3mm, respectively. In the validation with patient 4DCT dataset, the mean RMSE, NRMSE, SSIM, and three-dimensional tumor localization error in six patients were 73.7±19.6 HU, 9.2±2.6%, 0.88±0.04, and 0.8±0.6mm, respectively. These results suggest that the accuracy of the proposed technique is adequate when the respiratory motion is within the range of the training dataset. In the evaluation with a marker displacement larger than that of the training dataset, the mean RMSE, NRMSE, and tumor localization error were about 100 HU, 13%, and <2.0mm, respectively, except for one case having large motion variation. The performance of the proposed method was similar to those of previous studies. Processing time to generate the volumetric image was <100ms. We have shown the feasibility of the real-time CT image generation technique for volumetric imaging.

A comparison between Asians and Caucasians in the dimensions of the femoral isthmus based on a 3D-CT analysis of 1189 adult femurs

IntroductionFor successful intramedullary implant placement at the femur, such as nailing in unstable proximal femur fractures, the use of an implant that at least reaches or exceeds the femoral isthmus and yields sufficient thickness is recommended. A number of complications after intramedullary femoral nailing have been reported, particularly in Asians. To understand the anatomical features of the proximal femur and their ethnic differences, we aimed to accurately calculate the femoral isthmus dimensions and proximal distance of Asians and Caucasians.MethodsIn total, 1189 Asian and Caucasian segmented 3D CT data sets of femurs were analyzed. The individual femoral isthmus diameter was precisely computed to investigate whether gender, femur length, age, ethnicity or body mass index have an influence on isthmus diameters.ResultsThe mean isthmus diameter of all femurs was 10.71 ± 2.2 mm. A significantly larger diameter was found in Asians when compared to Caucasians (p < 0.001). Age was a strong predictor of the isthmus diameter variability in females (p < 0.001, adjusted r2 = 0.299). With every year of life, the isthmus showed a widening of 0.08 mm in women. A Matched Pair Analysis of 150 female femurs showed a significant difference between isthmus diameter in Asian and Caucasian femurs (p = 0.05). In 50% of the cases the isthmus was found in a range of 2.4 cm between 16.9 and 19.3 cm distal to the tip of the greater trochanter. The female Asian femur differs from Caucasians as it is wider at the isthmus.ConclusionsIn absolute values, the proximal isthmus distance did not show much variation but is more proximal in Asians. The detailed data presented may be helpful in the development of future implant designs. The length and thickness of future standard implants may be considered based on the findings.

Motion management with variable cycle-based respiratory guidance method for carbon-ion pencil beam scanning treatment.

A novel variable cycle-based respiratory guidance method was proposed to synchronize the patterns between patients' breathing and the magnetic excitation of synchrotron under the mode of full-energy depth scanning beam delivery, in order to improve the treatment precision and efficiency for carbon ion therapy. Audio-visual biofeedback system with variable cycle-based respiratory guidance method was developed. We enrolled 6 healthy volunteers and a simulation study of the fixed cycle-based and variable cycle-based respiratory guidance with three treatment fractions was performed. A total of 72 breathing curves were collected for 4D dose calculations with three 4DCT datasets of lung tumor cases. Target dose coverage (D95: the percent dose covering 95% of the target), dose homogeneity (D5-D95), and treatment time were analyzed. The Wilcoxon signed-rank test was used for statistical difference analysis, and p<0.05 was considered significant. With the variable cycle-based respiratory guidance method, the breath hold phase of breathing curve could be synchronized with the synchrotron flat-top phase over time. The dose homogeneity was improved by factors of 1.94-2.92 compared to the fixed cycle-based respiratory guidance maneuvers alone or in combination with gating technique. Moreover, the treatment efficiency increased by 11-23%, depending on the duty cycle settings of the gating window. The proposed variable cycle-based respiratory guidance method could improve both the treatment efficiency and precision under the mode of the full-energy depth scanning beam delivery for synchrotron-based carbon ion therapy.

Bidirectional RNN-based Few Shot Learning for 3D Medical Image Segmentation

Segmentation of organs of interest in 3D medical images is necessary for accurate diagnosis and longitudinal studies. Though recent advances using deep learning have shown success for many segmentation tasks, large datasets are required for high performance and the annotation process is both time consuming and labor intensive. In this paper, we propose a 3D few shot segmentation framework for accurate organ segmentation using limited training samples of the target organ annotation. To achieve this, a U-Net like network is designed to predict segmentation by learning the relationship between 2D slices of support data and a query image, including a bidirectional gated recurrent unit (GRU) that learns consistency of encoded features between adjacent slices. Also, we introduce a transfer learning method to adapt the characteristics of the target image and organ by updating the model before testing with arbitrary support and query data sampled from the support data. We evaluate our proposed model using three 3D CT datasets with annotations of different organs. Our model yielded significantly improved performance over state-of-the-art few shot segmentation models and was comparable to a fully supervised model trained with more target training data.