4DCT Datasets Research Articles

Dosimetric Planning Comparison for Left Ventricle Avoidance in Non-small Cell Lung Cancer Radiotherapy.

Radiation may unintentionally injure myocardial tissue, potentially leading to radiation-induced cardiac disease (RICD), with the net benefit of non-small cell lung cancer (NSCLC) radiotherapy (RT) due to the proximity of the lung and heart. RTOG-0617 showed a greater reduction in overall survival (OS) comparing higher doses to standard radiation doses in NSCLC RT. V5GyHeart has been reported as an OS predictor in the first- and fifth-year follow-ups. A worsening OS trend was reported in another study where the mean left ventricle dose (mean LV) was ≥14.5 Gy. It is therefore important to spare the heart, specifically the LV, from radiation. Furthermore, dose-limiting factors toward the normal lung should be accounted for to prevent radiation-induced lung injury. The LV and left anterior descending artery (LAD) were also contoured on the average four-dimensional computed tomography (4D-CT) dataset that contained clinically defined targets and normal structures for stage III NSCLC RT. The prescribed treatment plans (n=15) were retrospectively optimized with the clinical goals of minimizing the mean LV and mean heart dose while maintaining the dose constraint of V20GyLung ≤30% and V95%PTV ≥95%. Dose-volume histograms were used to compare the heart and lung dosimetric parameters between the delivered and reoptimized RT plans. A significant reduction (p≤0.044) was observed in the mean LV, mean heart dose, mean LAD dose, max LADdose, and V5GyHeart from the reoptimized RT plans. V20GyLung ≤30% and V95%PTV ≥95% were maintained, and no differences were observed in the mean lung, V5GyLung, V20GyLung, mean esophagus, and max cord. Minimizing the LV dose in NSCLC RT plans is achievable and dosimetrically advantageous for the heart while maintaining dose constraints to the normal lung and maximizing tumor control. Radiation dose reduction to cardiac substructures may decrease the RICD risk in NSCLC patients.

Impact of breathing signal-guided dose modulation on step-and-shoot 4DCT image reconstruction.

Breathing signal-guided 4DCT sequence scanning such as the intelligent 4DCT (i4DCT) approach reduces imaging artifacts compared to conventional 4DCT. By design, i4DCT captures entire breathing cycles during beam-on periods, leading to redundant projection data and increased radiation exposure to patients exhibiting prolonged exhalation phases. A recently proposed breathing-guided dose modulation (DM) algorithm promises to lower the imaging dose by temporarily reducing the CT tube current, but the impact on image reconstruction and the resulting images have not beeninvestigated. We evaluate the impact of breathing signal-guided DM on 4DCT image reconstruction and correspondingimages. This study is designed as a comparative and retrospective analysis based on 104 4DCT datasets. Each dataset underwent retrospective reconstruction twice: (a) utilizing the acquired clinical projection data for reconstruction, which yields reference image data, and (b) excluding projections acquired during potential DM phases from image reconstruction, resulting in DM-affected image data. Resulting images underwent automatic organ segmentation (lung/liver). (Dis)Similarity of reference and DM-affected images were quantified by the Dice coefficient of the entire organ masks and the organ overlaps within the DM-affected slices. Further, for lung cases, (a) and (b) were deformably registered and median magnitudes of the obtained displacement field were computed. Eventually, for 17 lung cases, gross tumor volumes (GTV) were recontoured on both (a) and (b). Target volume similarity was quantified by the Hausdorffdistance. DM resulted in a median imaging dose reduction of 15.4% (interquartile range [IQR]: 11.3%-19.9%) for the present patient cohort. Dice coefficients for lung ( ) and liver ( ) patients were consistently high for both the entire organs and the DM-affected slices (IQR lung: [entire lung/DM-affected slices only] to ; IQR liver: to ), demonstrating that DM did not cause organ distortions or alterations. Median displacements for DM-affected to reference image registration varied; however, only two out of 73 cases exhibited a median displacement larger than one isotropic 1 voxel size. The impact on GTV definition for the end-exhalation phase was also minor (median Hausdorff distance: 0.38mm, IQR: 0.15-0.46mm). This study demonstrates that breathing signal-guided DM has a minimal impact on image reconstruction and image appearance while improving patient safety by reducing doseexposure.

Dosimetric comparison of proton therapy and CyberKnife in stereotactic body radiation therapy for liver cancers

Stereotactic body radiation therapy (SBRT) has been increasingly used for the ablation of liver tumours. CyberKnife and proton beam therapy (PBT) are two advanced treatment technologies suitable to deliver SBRT with high dose conformity and steep dose gradients. However, there is very limited data comparing the dosimetric characteristics of CyberKnife to PBT for liver SBRT. PBT and CyberKnife plans were retrospectively generated using 4DCT datasets of ten patients who were previously treated for hepatocellular carcinoma (HCC, N = 5) and liver metastasis (N = 5). Dose volume histogram data was assessed and compared against selected criteria; given a dose prescription of 54 Gy in 3 fractions for liver metastases and 45 Gy in 3 fractions for HCC, with previously published consensus-based normal tissue dose constraints. Comparison of evaluation parameters showed a statistically significant difference for target volume coverage and liver, lungs and spinal cord (p < 0.05) dose, while chest wall and skin did not indicate a significant difference between the two modalities. A number of optimal normal tissue constraints was violated by both the CyberKnife and proton plans for the same patients due to proximity of tumour to chest wall. PBT resulted in greater organ sparing, the extent of which was mainly dependent on tumour location. Tumours located on the liver periphery experienced the largest increase in organ sparing. Organ sparing for CyberKnife was comparable with PBT for small target volumes.

A back propagation neural network based respiratory motion modelling method.

This study presents the development of a backpropagation neural network-based respiratory motion modelling method (BP-RMM) for precisely tracking arbitrary points within lung tissue throughout free respiration, encompassing deep inspiration and expiration phases. Internal and external respiratory data from four-dimensional computed tomography (4DCT) are processed using various artificial intelligence algorithms. Data augmentation through polynomial interpolation is employed to enhance dataset robustness. A BP neural network is then constructed to comprehensively track lung tissue movement. The BP-RMM demonstrates promising accuracy. In cases from the public 4DCT dataset, the average target registration error (TRE) between authentic deep respiration phases and those forecasted by BP-RMM for 75 marked points is 1.819mm. Notably, TRE for normal respiration phases is significantly lower, with a minimum error of 0.511mm. The proposed method is validated for its high accuracy and robustness, establishing it as a promising tool for surgical navigation within the lung.

Evaluation of Reliability of Dynamic Scapholunate Distance Measured on 4D CT-Acquired Images

Abstract Purpose A technique to measure scapholunate distance based on four-dimensional computed tomography (4D CT)-acquired images is presented. Methods Intra-observer variability was evaluated through a repeated-measures study. A 4D CT of seven patients suspected of scapholunate lesion was performed. Anatomical landmarks were identified on a three-dimensional reconstructed model of the wrist. All 4D CT datasets were evaluated thrice by two observers. Standard deviation of the differences between two measurements, interclass correlation coefficient (ICC), standard error of measurement (SEM), and minimal detectable change (MDC) were calculated. Results Intra-observer variability for the expert observer (ICC &gt; 0.95) was lower than that of the novice observer (ICC &gt; 0.77) and interobserver variability was low (ICC &gt; 0.85). For the expert observer, measurement error (SEM &lt; 0.13 mm and MDC &lt; 0.36 mm) was smaller than that of the novice observer (SEM &lt; 0.45 mm and MDC &lt; 1.24 mm). Both SEM and MDC values were low, compared to the scan resolution and the absolute value of intervals. Conclusion The proposed assessment results in a reproducible and reliable measurement of scapholunate distance.

The impact of motion on onboard MRI-guided pencil beam scanned proton therapy treatments

Objective. Online magnetic resonance imaging (MRI) guidance could be especially beneficial for pencil beam scanned (PBS) proton therapy of tumours affected by respiratory motion. For the first time to our knowledge, we investigate the dosimetric impact of respiratory motion on MRI-guided proton therapy compared to the scenario without magnetic field. Approach. A previously developed analytical proton dose calculation algorithm accounting for perpendicular magnetic fields was extended to enable 4D dose calculations. For two geometrical phantoms and three liver and two lung patient cases, static treatment plans were optimised with and without magnetic field (0, 0.5 and 1.5 T). Furthermore, plans were optimised using gantry angle corrections (0.5 T +5° and 1.5 T +15°) to reproduce similar beam trajectories compared to the 0 T reference plans. The effect of motion was then considered using 4D dose calculations without any motion mitigation and simulating 8-times volumetric rescanning, with motion for the patient cases provided by 4DCT(MRI) data sets. Each 4D dose calculation was performed for different starting phases and the CTV dose coverage V 95% and homogeneity D 5%–D 95% were analysed. Main results. For the geometrical phantoms with rigid motion perpendicular to the beam and parallel to the magnetic field, a comparable dosimetric effect was observed independent of the magnetic field. Also for the five 4DCT(MRI) cases, the influence of motion was comparable for all magnetic field strengths with and without gantry angle correction. On average, the motion-induced decrease in CTV V 95% from the static plan was 17.0% and 18.9% for 1.5 T and 0.5 T, respectively, and 19.9% without magnetic field. Significance. For the first time, this study investigates the combined impact of magnetic fields and respiratory motion on MR-guided proton therapy. The comparable dosimetric effects irrespective of magnetic field strength indicate that the effects of motion for future MR-guided proton therapy may not be worse than for conventional PBS proton therapy.

Aorta Segmentation in 3D CT Images by Combining Image Processing and Machine Learning Techniques.

Aorta segmentation is extremely useful in clinical practice, allowing the diagnosis of numerous pathologies, such as dissections, aneurysms and occlusive disease. In such cases, image segmentation is prerequisite for applying diagnostic algorithms, which in turn allow the prediction of possible complications and enable risk assessment, which is crucial in saving lives. The aim of this paper is to present a novel fully automatic 3D segmentation method, which combines basic image processing techniques and more advanced machine learning algorithms, for detecting and modelling the aorta in 3D CT imaging data. An initial intensity threshold-based segmentation procedure is followed by a classification-based segmentation approach, based on a Markov Random Field network. The result of the proposed two-stage segmentation process is modelled and visualized. The proposed methodology was applied to 16 3D CT data sets and the extracted aortic segments were reconstructed as 3D models. The performance of segmentation was evaluated both qualitatively and quantitatively against other commonly used segmentation techniques, in terms of the accuracy achieved, compared to the actual aorta, which was defined manually by experts. The proposed methodology achieved superior segmentation performance, compared to all compared segmentation techniques, in terms of the accuracy of the extracted 3D aortic model. Therefore, the proposed segmentation scheme could be used in clinical practice, such as in treatment planning and assessment, as it can speed up the evaluation of the medical imaging data, which is commonly a lengthy and tedious process.

Deep learning-based conditional inpainting for restoration of artifact-affected 4D CT images.

4D CT imaging is an essential component of radiotherapy of thoracic and abdominal tumors. 4D CT images are, however, often affected by artifacts that compromise treatment planning quality and image informationreliability. In this work, deep learning (DL)-based conditional inpainting is proposed to restore anatomically correct image information of artifact-affectedareas. The restoration approach consists of a two-stage process: DL-based detection of common interpolation (INT) and double structure (DS) artifacts, followed by conditional inpainting applied to the artifact areas. In this context, conditional refers to a guidance of the inpainting process by patient-specific image data to ensure anatomically reliable results. The study is based on 65 in-house 4D CT images of lung cancer patients (48 with only slight artifacts, 17 with pronounced artifacts) and two publicly available 4D CT data sets that serve as independent external test sets. Automated artifact detection revealed a ROC-AUC of 0.99 for INT and of 0.97 for DS artifacts (in-house data). The proposed inpainting method decreased the average root mean squared error (RMSE) by 52% (INT) and 59% (DS) for the in-house data. For the external test data sets, the RMSE improvement is similar (50% and 59%, respectively). Applied to 4D CT data with pronounced artifacts (not part of the training set), 72% of the detectable artifacts were removed. The results highlight the potential of DL-based inpainting for restoration of artifact-affected 4D CT data. Compared to recent 4D CT inpainting and restoration approaches, the proposed methodology illustrates the advantages of exploiting patient-specific prior imageinformation.

ORRN: An ODE-Based Recursive Registration Network for Deformable Respiratory Motion Estimation With Lung 4DCT Images.

Deformable Image Registration (DIR) plays a significant role in quantifying deformation in medical data. Recent Deep Learning methods have shown promising accuracy and speedup for registering a pair of medical images. However, in 4D (3D + time) medical data, organ motion, such as respiratory motion and heart beating, can not be effectively modeled by pair-wise methods as they were optimized for image pairs but did not consider the organ motion patterns necessary when considering 4D data. This article presents ORRN, an Ordinary Differential Equations (ODE)-based recursive image registration network. Our network learns to estimate time-varying voxel velocities for an ODE that models deformation in 4D image data. It adopts a recursive registration strategy to progressively estimate a deformation field through ODE integration of voxel velocities. We evaluate the proposed method on two publicly available lung 4DCT datasets, DIRLab and CREATIS, for two tasks: 1) registering all images to the extreme inhale image for 3D+t deformation tracking and 2) registering extreme exhale to inhale phase images. Our method outperforms other learning-based methods in both tasks, producing the smallest Target Registration Error of 1.24 mm and 1.26 mm, respectively. Additionally, it produces less than 0.001% unrealistic image folding, and the computation speed is less than 1 s for each CT volume. ORRN demonstrates promising registration accuracy, deformation plausibility, and computation efficiency on group-wise and pair-wise registration tasks. It has significant implications in enabling fast and accurate respiratory motion estimation for treatment planning in radiation therapy or robot motion planning in thoracic needle insertion.

Continuous time-resolved estimated synthetic 4D-CTs for dose reconstruction of lung tumor treatments at a 0.35 T MR-linac

Objective. To experimentally validate a method to create continuous time-resolved estimated synthetic 4D-computed tomography datasets (tresCTs) based on orthogonal cine MRI data for lung cancer treatments at a magnetic resonance imaging (MRI) guided linear accelerator (MR-linac). Approach. A breathing porcine lung phantom was scanned at a CT scanner and 0.35 T MR-linac. Orthogonal cine MRI series (sagittal/coronal orientation) at 7.3 Hz, intersecting tumor-mimicking gelatin nodules, were deformably registered to mid-exhale 3D-CT and 3D-MRI datasets. The time-resolved deformation vector fields were extrapolated to 3D and applied to a reference synthetic 3D-CT image (sCTref), while accounting for breathing phase-dependent lung density variations, to create 82 s long tresCTs at 3.65 Hz. Ten tresCTs were created for ten tracked nodules with different motion patterns in two lungs. For each dataset, a treatment plan was created on the mid-exhale phase of a measured ground truth (GT) respiratory-correlated 4D-CT dataset with the tracked nodule as gross tumor volume (GTV). Each plan was recalculated on the GT 4D-CT, randomly sampled tresCT, and static sCTref images. Dose distributions for corresponding breathing phases were compared in gamma (2%/2 mm) and dose–volume histogram (DVH) parameter analyses. Main results. The mean gamma pass rate between all tresCT and GT 4D-CT dose distributions was 98.6%. The mean absolute relative deviations of the tresCT with respect to GT DVH parameters were 1.9%, 1.0%, and 1.4% for the GTV D 98%, D 50%, and D 2%, respectively, 1.0% for the remaining nodules D 50%, and 1.5% for the lung V 20Gy. The gamma pass rate for the tresCTs was significantly larger (p < 0.01), and the GTV D 50% deviations with respect to the GT were significantly smaller (p < 0.01) than for the sCTref. Significance. The results suggest that tresCTs could be valuable for time-resolved reconstruction and intrafractional accumulation of the dose to the GTV for lung cancer patients treated at MR-linacs in the future.

An approach to generate synthetic 4DCT datasets to benchmark Mid-Position implementations

PurposeThe Mid-Position image is constructed from 4DCT data using Deformable Image Registration and can be used as planning CT with reduced PTV volumes. 4DCT datasets currently-available for testing do not provide the corresponding Mid-P images of the datasets. This work describes an approach to generate human-like synthetic 4DCT datasets with the associated Mid-P images that can be used as reference in the validation of Mid-P implementations. MethodsTwenty synthetic 4DCT datasets with the associated reference Mid-P images were generated from twenty clinical 4DCT datasets. Per clinical dataset, an anchor phase was registered to the remaining nine phases to obtain nine Deformable Vector Fields (DVFs). These DVFs were used to warp the anchor phase in order to generate the synthetic 4DCT dataset and the corresponding reference Mid-P image. Similarly, a reference 4D tumor mask dataset and its corresponding Mid-P tumor mask were generated. The generated synthetic datasets and masks were used to compare and benchmark the outcomes of three independent Mid-P implementations using a set of experiments. ResultsThe Mid-P images constructed by the three implementations showed high similarity scores when compared to the reference Mid-P images except for one noisy dataset. The biggest difference in the estimated motion amplitudes (-2.6 mm) was noticed in the Superior-Inferior direction. The statistical analysis showed no significant differences among the three implementations for all experiments. ConclusionThe described approach and the proposed experiments provide an independent method that can be used in the validation of any Mid-P implementation being developed.

Heart Dose Uncertainty Caused by Heartbeat during Adjuvant Breast Radiotherapy

Adjuvant radiotherapy is a crucial treatment option for breast cancer. Owing to the long survival after treatment, heart toxicity has become a critical issue in those who receive radiotherapy. The heart dose is calculated using the computed tomography (CT) images acquired during simulation. The heart volume is quite different between the systolic and diastolic phases. This study used four-dimensional (4D) CT to capture the heart image and determine the ranges of the heart dose in different phases of the heart. Patients with left-sided breast cancer who underwent breast conservative surgery followed by adjuvant radiotherapy to the entire breast between January 2016 and December 2021 were identified. The three-dimensional (3D) CT and 4DCT datasets of the patients were acquired to evaluate the variation in the heart dose. The heart volume in each respiratory phase was depicted in a box plot along with the dice similarity coefficient (DCS). A total of 32 patients met the inclusion criteria. The DCS of the heart in each respiration phase was within the range of 0.91 to 0.95. Variations in the heart volume and morphology were not evident. However, these variations were not influenced by the respiratory cycle. The median relative variations in the 4DCT volume, mean dose, V20 (≥20 Gy), and V40 (40 Gy) were 6%, 14%, 19%, and 32%, respectively. All parameters showed significant variation between the maximum values in 4DCT and planning 3DCT. When parameters in planning 3DCT were compared to minimum values in 4DCT, the heart volume and V20 still showed significant differences, whereas the mean dose (p = .052) and V40 (p = .151) did not. We found that the median variation in the heart volume during the respiratory cycle was 6%, which led to a 14% variation in the mean heart dose, 19% in V20, and 32% in V40. Dose variation should not be ignored. A more precise method for estimating the heart dose during treatment should be investigated.

Deformable lung 4DCT image registration via landmark-driven cycle network.

An automated, accurate, and efficient lung four-dimensional computed tomography (4DCT) image registration method is clinically important to quantify respiratory motion for optimal motion management. The purpose of this work is to develop a weakly supervised deep learning method for 4DCT lung deformable image registration (DIR). The landmark-driven cycle network is proposed as a deep learning platform that performs DIR of individual phase datasets in a simulation 4DCT. This proposed network comprises a generator and a discriminator. The generator accepts moving and target CTs as input and outputs the deformation vector fields (DVFs) to match the two CTs. It is optimized during both forward and backward paths to enhance the bi-directionality of DVF generation. Further, the landmarks are used to weakly supervise the generator network. Landmark-driven loss is used to guide the generator's training. The discriminator then judges the realism of the deformed CT to provide extra DVF regularization. We performed four-fold cross-validation on 10 4DCT datasets from the public DIR-Lab dataset and a hold-out test on our clinic dataset, which included 50 4DCT datasets. The DIR-Lab dataset was used to evaluate the performance of the proposed method against other methods in the literature by calculating the DIR-Lab Target Registration Error (TRE). The proposed method outperformed other deep learning-based methods on the DIR-Lab datasets in terms of TRE. Bi-directional and landmark-driven loss were shown to be effective for obtaining high registration accuracy. The mean and standard deviation of TRE for the DIR-Lab datasets was 1.20±0.72mm and the mean absolute error (MAE) and structural similarity index (SSIM) for our datasets were 32.1±11.6 HU and 0.979±0.011, respectively. The landmark-driven cycle network has been validated and tested for automatic deformable image registration of patients' lung 4DCTs with results comparable to or better than competing methods.

Shallow kinetics induced by a metronome (SKIM): A novel contactless respiratory motion management.

As an alternative to conventional compression amidst the COVID-19 pandemic, we developed a contactless motion management strategy. By increasing the patient's breathing rate to induce shallow breathing with the aid of a metronome, our hypothesis is that the motion magnitude of the target may be minimized without physical contact or compression. Fourteen lung stereotactic body radiation therapy (SBRT) patients treated under fast shallow-breathing (FSB) were selected for inclusion in this retrospective study. Our proposed method is called shallow kinetics induced by a metronome (SKIM). We induce FSB by setting the beats-per-minute (BPM) high (typically in the range of 50-60). This corresponded to a patient breathing rate of 25-30 (breathing) cycles per minute. The magnitude of target motion in 3D under SKIM was evaluated using 4DCT datasets. Comparison with free breathing (FB) 4DCT was also made for a subset for which FB data available. The overall effectiveness of SKIM was evaluated with 18 targets (14 patients). Direct comparison with FB was performed with 12 targets (10patients). The vector norm mean±SD value of motion magnitude under SKIM for 18 targets was 8.2±4.1mm. The mean±SD metronome BPM was 54.9±4.0 in this group. The vector norm means±SD values of target motion for FB and SKIM in the 12 target sub-group were 14.6±8.5mm and 9.3±3.7mm, respectively. The mean±SD metronome BPM for this sub-group was 56.3±2.5. Compared with FB, SKIM can significantly reduce respiratory motion magnitude of thoracic targets. The difference in maximum motion reduction in the overall vector norm, S-I, and A-P directions was significant (p=0.033, 0.042, 0.011). Our proposed method can be an excellent practical alternative to conventional compression due to its flexibility and ease of implementation.

Experimental validation of a 4D dynamic dose calculation model for proton pencil beam scanning without spot time stamp considering free-breathing motion.

We developed a 4-dimensional dynamic dose (4DDD) calculation model for proton pencil beam scanning (PBS). This model incorporates the spill start time for all energies and uses the remaining irradiated spot time model instead of irradiated spot time logs. This study aimed to validate the calculation accuracy of a log file-based 4DDD model by comparing it with dose measurements performed under free-breathing conditions, thereby serving as an alternative approach to the conventional log file-based system. Three cubic verification plans were created using a heterogeneous block phantom; these plans were created using 10 phase 4D-CT datasets of the phantom. The CIRS dynamic platform was used to simulate motion with amplitudes of 2.5, 3.75, and 5.0mm. These plans consisted of eight- and two-layered rescanning techniques. The lateral profiles were measured using a 2D ionization chamber array (2D-array) and EBT3 Gafchromic films at four starting phases, including three sinusoidal curves (periods of 3, 4, and 6s) and a representative patient curve during actual treatment. 4DDDs were calculated using in-house scripting that assigned a time stamp to each spot and performed dose accumulation using deformable image registration. Furthermore, to evaluate the impact of parameter selection on our 4DDD model calculations, simulations were performed assuming a ±10% change in irradiation time stamp (0.8±0.08s) and spot scan speed. We evaluated the 2D gamma index and the absolute point doses between the calculated values and the measurements. The 2D-array measurements revealed that the gamma scores for the static plans (no motion) and 4DDD plans exceeded 97.5% and 93.9% at 3%/3mm, respectively. The average gamma score of the 4DDD plans was at least 96.1%. When using EBT3 films, the gamma scores of the 4DDD model exceeded 92.4% and 98.7% at 2%/2mm and 3%/3mm, respectively. Regarding the 4DDD point dose differences, more than 95% of the dose regions exhibited discrepancies within ±5.0% for 97.7% of the total points across all plans. The spot time assignment accuracy of our 4DDD model was acceptable even with ±10% sensitivity. However, the accuracy of the scan speed, when varied within ±10%sensitivity, was not acceptable (minimum gamma scores of 82.6% and maximum dose difference of 12.9%). Our 4DDD calculations under free-breathing conditions using amplitudes of less than 5.0mm were in good agreement with the measurements regardless of the starting phases, breathing curve patterns (between 3 and 6s periods), and varying numbers of layered rescanning. The proposed system allows us to evaluate actual irradiated doses in various breathing periods, amplitudes, and starting phases, even on PBS machines without the ability to record spot logs.

4D-CT deformable image registration using unsupervised recursive cascaded full-resolution residual networks.

A novel recursive cascaded full-resolution residual network (RCFRR-Net) for abdominal four-dimensional computed tomography (4D-CT) image registration was proposed. The entire network was end-to-end and trained in the unsupervised approach, which meant that the deformation vector field, which presented the ground truth, was not needed during training. The network was designed by cascading three full-resolution residual subnetworks with different architectures. The training loss consisted of the image similarity loss and the deformation vector field regularization loss, which were calculated based on the final warped image and the fixed image, allowing all cascades to be trained jointly and perform the progressive registration cooperatively. Extensive network testing was conducted using diverse datasets, including an internal 4D-CT dataset, a public DIRLAB 4D-CT dataset, and a 4D cone-beam CT (4D-CBCT) dataset. Compared with the iteration-based demon method and two deep learning-based methods (VoxelMorph and recursive cascaded network), the RCFRR-Net achieved consistent and significant gains, which demonstrated that the proposed method had superior performance and generalization capability in medical image registration. The proposed RCFRR-Net was a promising tool for various clinical applications.

Frequency constraint-based adversarial attack on deep neural networks for medical image classification

The security of AI systems has gained significant attention in recent years, particularly in the medical diagnosis field. To develop a secure medical image classification system based on deep neural networks, it is crucial to design effective adversarial attacks that can embed hidden, malicious behaviors into the system. However, designing a unified attack method that can generate imperceptible attack samples with high content similarity and be applied to diverse medical image classification systems is challenging due to the diversity of medical imaging modalities and dimensionalities. Most existing attack methods are designed to attack natural image classification models, which inevitably corrupt the semantics of pixels by applying spatial perturbations. To address this issue, we propose a novel frequency constraint-based adversarial attack method capable of delivering attacks in various medical image classification tasks. Specially, our method introduces a frequency constraint to inject perturbation into high-frequency information while preserving low-frequency information to ensure content similarity. Our experiments include four public medical image datasets, including a 3D CT dataset, a 2D chest X-Ray image dataset, a 2D breast ultrasound dataset, and a 2D thyroid ultrasound dataset, which contain different imaging modalities and dimensionalities. The results demonstrate the superior performance of our model over other state-of-the-art adversarial attack methods for attacking medical image classification tasks on different imaging modalities and dimensionalities.

Three-Dimensional Echocardiography Derived Printing: A Review of Workflow, Current, and Future Applications

While most of the work pertaining to 3D printing in cardiology has been based on 3D CT and MRI datasets, due to the recent advents in 3D printing and 3D echocardiography, 3D printing from echocardiography has now become feasible. In this review, we discuss the workflow, applications, limitations, and potential future directions of 3D echocardiography-based 3D printing. 3D printing using 3D echocardiographic datasets has been successfully deployed in the field of cardiovascular medicine in the twenty-first century. It was shown to provide a significant additive value particularly with regards to visualization of valves and subvalvar apparatus. Given its limitations of limited field of view and somewhat lower spatial resolution, this approach is likely ideally suited for a combined multi-modality-based printing. Patient-specific, 3D echocardiography-derived 3D-printed models are now quite feasible. This can be used for surgical/procedural planning and training for optimal outcomes.

Deformable registration of lung 3DCT images using an unsupervised heterogeneous multi-resolution neural network.

Lung image registration is more challenging than other organs. This is because the breath of the human body causes large deformations in the lung parenchyma and small deformations in tissues such as the pulmonary vascular. Many studies have recently used multi-resolution networks to solve the lung registration problem. However, they use the same structure of registration modules on each level, which makes it difficult to handle complex and small deformations. We propose an unsupervised heterogeneous multi-resolution network (UHMR-Net) to overcome the above problem. The image detail registration module (IDRM) is designed on the highest resolution level. Within this module, the cascaded network is used on the same resolution image to continuously learn the "remaining" detail deformation fields. The shallow shrinkage loss (SS-Loss) is designed to supervise the cascaded network, thus further improving the ability of the network to handle small deformations. Moreover, with the lightweight feature local correlation layer we proposed, the image boundary registration module (IBRM), on multiple low-resolution levels, can better solve the large deformation registration problem. The target registration error on the public DIR-Lab 4DCT dataset was 1.56 ± 1.39mm, which was significantly better than the classic conventional methods and advanced deep-based methods.

Light Pareto robust optimization for IMRT treatment planning.

Robust optimization (RO) has been proposed to mitigate breathing motion uncertainty during treatment in intensity-modulated radiation therapy (IMRT) planning for breast or lung cancer. RO is a pessimistic approach that implicitly trades off average-case for worst-case treatment plan quality. Pareto robust optimization (PRO) provides a mechanism for improving nonworst-case plan outcomes, but often remains overly conservative in the averagecase. The goal of this study is to characterize the trade-off between the optimality of robust IMRT plans in the worst case and the treatment quality in nonworst-case realizations of breathing motion. We provide a light Pareto robust optimization (LPRO) method for IMRT and test its clinical viability for improving the average-case plan quality while preserving robustness, in comparison to RO and PRO plans. Five clinical left-sided breast cancer patients were included in the study, each with an associated 4D-CT dataset approximating their breathing cycle. Using simulation, 50 different breathing patterns were generated for each patient. A first-stage optimization was solved with the objective of cardiac sparing while ensuring robustness on the target dose under breathing uncertainty. Next, a second-stage objective of overdose minimization was considered to improve plan quality in a controlled LPRO framework. For the simulated breathing scenarios, the trade-off between loss of average cardiac sparing at worst-case and the overdose to the breast was quantified by calculating the accumulated dose for each plan in each breathing scenario. Finally, the RO, PRO, and LPRO plans were each evaluated using eight clinical dose-volume criteria on the target and organs atrisk. The LPRO models allowed for significantly sharper dose falloffs in the expected dose instances, relative to both RO and PRO models. Plans began looking valid for delivery with average allowances of as little as +0.1Gy additional dose to the heart, and most patients experienced diminishing returns beyond +0.2Gy. Without sacrificing robustness, the LPRO approach produces viable plans with true total-target irradiation. Furthermore, the plans produced were able to reduce the nonworst-case downside typical of RO, without the characteristic overdosing or average-case pessimism seen in prior models.