Registration Algorithm Research Articles

An AI-Based Workflow for Fast Registration of UAV-Produced 3D Point Clouds

The integration of structure from motion (SFM) and unmanned aerial vehicle (UAV) technologies has allowed for the generation of very high-resolution three-dimensional (3D) point cloud data (up to millimeters) to detect and monitor surface changes. However, a bottleneck still exists in accurately and rapidly registering the point clouds at different times. The existing point cloud registration algorithms, such as the Iterative Closest Point (ICP) and the Fast Global Registration (FGR) method, were mainly developed for the registration of small and static point cloud data, and do not perform well when dealing with large point cloud data with potential changes over time. In particular, registering large data is computationally expensive, and the inclusion of changing objects reduces the accuracy of the registration. In this paper, we develop an AI-based workflow to ensure high-quality registration of the point clouds generated using UAV-collected photos. We first detect stable objects from the ortho-photo produced by the same set of UAV-collected photos to segment the point clouds of these objects. Registration is then performed only on the partial data with these stable objects. The application of this workflow using the UAV data collected from three erosion plots at the East Tennessee Research and Education Center indicates that our workflow outperforms the existing algorithms in both computational speed and accuracy. This AI-based workflow significantly improves computational efficiency and avoids the impact of changing objects for the registration of large point cloud data.

A Practical Star Image Registration Algorithm Using Radial Module and Rotation Angle Features

Star image registration is the most important step in the application of astronomical image differencing, stacking, and mosaicking, which requires high robustness, accuracy, and real-time capability on the part of the algorithm. At present, there are no high-performance registration algorithms available in this field. In the present paper, we propose a star image registration algorithm that relies only on radial module features (RMF) and rotation angle features (RAF) while providing excellent robustness, high accuracy, and good real-time performance. The test results on a large amount of simulated and real data show that the comprehensive performance of the proposed algorithm is significantly better than the four classical baseline algorithms as judged by the presence of rotation, insufficient overlapping area, false stars, position deviation, magnitude deviation, and complex sky background, making it a more ideal star image registration algorithm than current alternatives.

Deep learning-assisted construction of three-dimensional face midsagittal plane based on point clouds

Objective: To establish an intelligent registration algorithm under the framework of original-mirror alignment algorithm to construct three-dimensional(3D) facial midsagittal plane automatically. Dynamic Graph Registration Network (DGRNet) was established to realize the intelligent registration, in order to provide a reference for clinical digital design and analysis. Methods: Two hundred clinical patients without significant facial deformities were collected from October 2020 to October 2022 at Peking University School and Hospital of Stomatology. The DGRNet consists of constructing the feature vectors of key points in point original and mirror point clouds (X, Y), obtaining the correspondence of key points, and calculating the rotation and translation by singular value decomposition. Original and mirror point clouds were registrated and united. The principal component analysis (PCA) algorithm was used to obtain the DGRNet alignment midsagittal plane. The model was evaluated based on the coefficient of determination (R2) index for the translation and rotation matrix of test set. The angle error was evaluated on the 3D facial midsagittal plane constructed by the DGRNet alignment midsagittal plane and the iterative closet point(ICP) alignment midsagittal plane for 50 cases of clinical facial data. Results: The average angle error of the DGRNet alignment midsagittal plane and ICP alignment midsagittal plane was 1.05°±0.56°, and the minimum angle error was only 0.13°. The successful detection rate was 78%(39/50) within 1.50° and 90% (45/50)within 2.00°. Conclusions: This study proposes a new solution for the construction of 3D facial midsagittal plane based on the DGRNet alignment method with intelligent registration, which can improve the efficiency and effectiveness of treatment to some extent.

Development and evaluation of robot-assisted ultrasound navigation system for pedicle screw placement: An ex-vivo animal validation.

Spinal instrumentation with pedicle screw placement (PSP) is an important surgical technique for spinal diseases. Accurate screw trajectory is a prerequisite for PSP. Ultrasound (US) imaging with robot-assisted system forms a non-radiative alternative to provide precise screw trajectory. This study reports on the development and assessment of US navigation for this application. A robot-assisted US reconstruction was proposed and an automatic CT-to-US registration algorithm was investigated, allowing the registration of screw trajectories. Experiments were conducted on ex-vivo lamb spines to evaluate system performance. In total, 72 screw trajectories are measured, displaying an average position accuracy of 2.80±1.14mm and orientation accuracy of 1.38±0.61°. The experimental results demonstrate the feasibility of proposed US system. This work, although restricted to laboratory settings, encourages further exploration of the potential of this technology in clinical practice.

Accurate Global Point Cloud Registration Using GPU-Based Parallel Angular Radon Spectrum.

Accurate robot localization and mapping can be improved through the adoption of globally optimal registration methods, like the Angular Radon Spectrum (ARS). In this paper, we present Cud-ARS, an efficient variant of the ARS algorithm for 2D registration designed for parallel execution of the most computationally expensive steps on Nvidia™ Graphics Processing Units (GPUs). Cud-ARS is able to compute the ARS in parallel blocks, with each associated to a subset of input points. We also propose a global branch-and-bound method for translation estimation. This novel parallel algorithm has been tested on multiple datasets. The proposed method is able to speed up the execution time by two orders of magnitude while obtaining more accurate results in rotation estimation than state-of-the-art correspondence-based algorithms. Our experiments also assess the potential of this novel approach in mapping applications, showing the contribution of GPU programming to efficient solutions of robotic tasks.

Partial point cloud registration algorithm based on deep learning and non-corresponding point estimation

Partial point cloud registration algorithm based on deep learning and non-corresponding point estimation

Fast Automatic Registration of UAV Images via Bidirectional Matching.

Image registration plays a vital role in the mosaic process of multiple UAV (Unmanned Aerial Vehicle) images acquired from different spatial positions of the same scene. Aimed at the problem that many fast registration methods cannot provide both high speed and accuracy simultaneously for UAV visible light images, this work proposes a novel registration framework based on a popular baseline registration algorithm, ORB-the Oriented FAST (Features from Accelerated Segment Test) and Rotated BRIEF (Binary Robust Independent Elemental Features) algorithm. First, the ORB algorithm is utilized to extract image feature points fast. On this basis, two bidirectional matching strategies are presented to match obtained feature points. Then, the PROSRC (Progressive Sample Consensus) algorithm is applied to remove false matches. Finally, the experiments are carried out on UAV image pairs about different scenes including urban, road, building, farmland, and forest. Compared with the original version and other state-of-the-art registration methods, the bi-matching ORB algorithm exhibits higher accuracy and faster speed without any training or prior knowledge. Meanwhile, its complexity is quite low for on-board realization.

Exploring the performance of implicit neural representations for brain image registration

Pairwise image registration is a necessary prerequisite for brain image comparison and data integration in neuroscience and radiology. In this work, we explore the efficacy of implicit neural representations (INRs) in improving the performance of brain image registration in magnetic resonance imaging. In this setting, INRs serve as a continuous and coordinate based approximation of the deformation field obtained through a multi-layer perceptron. Previous research has demonstrated that sinusoidal representation networks (SIRENs) surpass ReLU models in performance. In this study, we first broaden the range of activation functions to further investigate the registration performance of implicit networks equipped with activation functions that exhibit diverse oscillatory properties. Specifically, in addition to the SIRENs and ReLU, we evaluate activation functions based on snake, sine+, chirp and Morlet wavelet functions. Second, we conduct experiments to relate the hyper-parameters of the models to registration performance. Third, we propose and assess various techniques, including cycle consistency loss, ensembles and cascades of implicit networks, as well as a combined image fusion and registration objective, to enhance the performance of implicit registration networks beyond the standard approach. The investigated implicit methods are compared to the VoxelMorph convolutional neural network and to the symmetric image normalization (SyN) registration algorithm from the Advanced Normalization Tools (ANTs). Our findings not only highlight the remarkable capabilities of implicit networks in addressing pairwise image registration challenges, but also showcase their potential as a powerful and versatile off-the-shelf tool in the fields of neuroscience and radiology.

Prostate volume analysis in image registration for prostate cancer care: a verification study

Combined magnetic resonance imaging (MRI) and positron emission tomography/computed tomography (PET/CT) may enhance diagnosis, aid surgical planning and intra-operative orientation for prostate biopsy and radical prostatectomy. Although PET-MRI may provide these benefits, PET-MRI machines are not widely available. Image fusion of Prostate specific membrane antigen PET/CT and MRI acquired separately may be a suitable clinical alternative. This study compares CT-MR registration algorithms for urological prostate cancer care. Paired whole-pelvis MR and CT scan data were used (n = 20). A manual prostate CTV contour was performed independently on each patients MR and CT image. A semi-automated rigid-, automated rigid- and automated non-rigid registration technique was applied to align the MR and CT data. Dice Similarity Index (DSI), 95% Hausdorff distance (95%HD) and average surface distance (ASD) measures were used to assess the closeness of the manual and registered contours. The automated non-rigid approach had a significantly improved performance compared to the automated rigid- and semi-automated rigid-registration, having better average scores and decreased spread for the DSI, 95%HD and ASD (all p < 0.001). Additionally, the automated rigid approach had similar significantly improved performance compared to the semi-automated rigid registration across all accuracy metrics observed (all p < 0.001). Overall, all registration techniques studied here demonstrated sufficient accuracy for exploring their clinical use. While the fully automated non-rigid registration algorithm in the present study provided the most accurate registration, the semi-automated rigid registration is a quick, feasible, and accessible method to perform image registration for prostate cancer care by urologists and radiation oncologists now.

A novel mixed reality-guided dental implant placement navigation system based on virtual-actual registration

BackgroundsThe key to successful dental implant surgery is to place the implants accurately along the pre-operative planned paths. The application of surgical navigation systems can significantly improve the safety and accuracy of implantation. However, the frequent shift of the views of the surgeon between the surgical site and the computer screen causes troubles, which is expected to be solved by the introduction of mixed-reality technology through the wearing of HoloLens devices by enabling the alignment of the virtual three-dimensional (3D) image with the actual surgical site in the same field of view. MethodsThis study utilized mixed reality technology to enhance dental implant surgery navigation. Our first step was reconstructing a virtual 3D model from pre-operative cone-beam CT (CBCT) images. We then obtained the relative position between objects using the navigation device and HoloLens camera. Via the algorithms of virtual-actual registration, the transformation matrixes between the HoloLens devices and the navigation tracker were acquired through the HoloLens-tracker registration, and the transformation matrixes between the virtual model and the patient phantom through the image-phantom registration. In addition, the algorithm of surgical drill calibration assisted in acquiring transformation matrixes between the surgical drill and the patient phantom. These algorithms allow real-time tracking of the surgical drill's location and orientation relative to the patient phantom under the navigation device. With the aid of the HoloLens 2, virtual 3D images and actual patient phantoms can be aligned accurately, providing surgeons with a clear visualization of the implant path. ResultsPhantom experiments were conducted using 30 patient phantoms, with a total of 102 dental implants inserted. Comparisons between the actual implant paths and the pre-operatively planned implant paths showed that our system achieved a coronal deviation of 1.507 ± 0.155 mm, an apical deviation of 1.542 ± 0.143 mm, and an angular deviation of 3.468 ± 0.339°. The deviation was not significantly different from that of the navigation-guided dental implant placement but better than the freehand dental implant placement. ConclusionOur proposed system realizes the integration of the pre-operative planned dental implant paths and the patient phantom, which helps surgeons achieve adequate accuracy in traditional dental implant surgery. Furthermore, this system is expected to be applicable to animal and cadaveric experiments in further studies.

Hybrid unsupervised paradigm based deformable image fusion for 4D CT lung image modality

Deformable image registration plays a critical role in various clinical applications (e.g., image fusion, atlas creation, and tumors targeting). In radiation therapy, especially in the context of fast registration of computed tomography (CT) lung image modalities, it is used to determine the geometric transformation by relating the anatomic points in two images. The main challenge lies in effectively addressing the nonlinear large and small deformation between the inspiration and expiration phases. In this work, we propose an unsupervised hybrid paradigm-based registration network (HPRN) for the registration of 4D CT lung images without relying on ground truth data. The proposed HPRN exhibits effective learning of multi-scale and multi-resolution features, leading to the computation of a more accurate Deformation Vector Field (DVF). Furthermore, we incorporate the regularization, image similarity and Jacobian determinant loss functions, which results in improving capability in dealing with complex large and small deformations. We evaluate the effectiveness of the proposed model on the publicly accessible DIRLab 4DCT lung image dataset, which shows the effectiveness of the proposed framework by achieving better Target Registration Error (2.04±1.42mm) compared to other state-of-the-art unsupervised image registration algorithms.

EDIRNet: an unsupervised deformable registration model for X-ray and neutron images.

For high-precision industrial non-destructive testing, multimodal image registration technology can be employed to register X-ray and neutron images. X-ray and neutron image registration algorithms usually use conventional methods through iterative optimization. These methods will increase the cost of registration time and require more initialization parameters. The imaging results of internal sample structures can suffer from edge blurring due to the influence of a neutron beam collimator aperture, X-ray focal point, and imaging angles. We present an unsupervised learning model, EDIRNet, based on deep learning for deformable registration of X-ray and neutron images. We define the registration process as a function capable of estimating the flow field from input images. By leveraging deep learning techniques, we effectively parameterize this function. Consequently, given a registration image, our optimized network parameters enable rapid and direct estimation of the flow field between the images. We design an attention-based edge enhancement module to enhance the edge features of the image. For evaluating our presented network model, we utilize a dataset including 552 pairs of X-ray and neutron images. The experimental results show that the registration accuracy of EDIRNet reaches 93.09%. Compared with traditional algorithms, the accuracy of EDIRNet is improved by 3.17%, and the registration time is reduced by 28.75s.

Comparative analysis of alignment algorithms for macular optical coherence tomography imaging

BackgroundOptical coherence tomography (OCT) is the most important and commonly utilized imaging modality in ophthalmology and is especially crucial for the diagnosis and management of macular diseases. Each OCT volume is typically only available as a series of cross-sectional images (B-scans) that are accessible through proprietary software programs which accompany the OCT machines. To maximize the potential of OCT imaging for machine learning purposes, each OCT image should be analyzed en bloc as a 3D volume, which requires aligning all the cross-sectional images within a particular volume.MethodsA dataset of OCT B-scans obtained from 48 age-related macular degeneration (AMD) patients and 50 normal controls was used to evaluate five registration algorithms. After alignment of B-scans from each patient, an en face surface map was created to measure the registration quality, based on an automatically generated Laplace difference of the surface map–the smoother the surface map, the smaller the average Laplace difference. To demonstrate the usefulness of B-scan alignment, we trained a 3D convolutional neural network (CNN) to detect age-related macular degeneration (AMD) on OCT images and compared the performance of the model with and without B-scan alignment.ResultsThe mean Laplace difference of the surface map before registration was 27 ± 4.2 pixels for the AMD group and 26.6 ± 4 pixels for the control group. After alignment, the smoothness of the surface map was improved, with a mean Laplace difference of 5.5 ± 2.7 pixels for Advanced Normalization Tools Symmetric image Normalization (ANTs-SyN) registration algorithm in the AMD group and a mean Laplace difference of 4.3 ± 1.4.2 pixels for ANTs in the control group. Our 3D CNN achieved superior performance in detecting AMD, when aligned OCT B-scans were used (AUC 0.95 aligned vs. 0.89 unaligned).ConclusionsWe introduced a novel metric to quantify OCT B-scan alignment and compared the effectiveness of five alignment algorithms. We confirmed that alignment could be improved in a statistically significant manner with readily available alignment algorithms that are available to the public, and the ANTs algorithm provided the most robust performance overall. We further demonstrated that alignment of OCT B-scans will likely be useful for training 3D CNN models.

Modified maximum likelihood space registration method for shipborne multi-radar signal processing

Aiming at the spatial registration problem for networking of multi-platform shipborne radars, a modified maximum likelihood (MML) absolute registration algorithm based on singular value decomposition (SVD) is proposed. Firstly, the four major biases (radar detection biases, platform positioning biases, attitude biases, and baseline conversion biases) which affect shipborne radar detection are analyzed, and an equivalent model of shipborne radar detection biases is established to make the bias parameters more efficiently. Secondly, the maximum likelihood registration method is improved. According to the singular value decomposition, the conditional index is used to evaluate the observability in the maximum likelihood registration method in realtime, and the abnormal conditional index is corrected to effectively reduce the fluctuation of the bias estimation. Finally, the simulation results of two different scenarios are conducted, and the results show that the new method has significant improvements in estimation accuracy and algorithm stability compared with the existing algorithms.

Dosimetric Parameters Predictive of Treatment-Related Toxicity in High Dose-Rate Brachytherapy as Monotherapy for Prostate Cancer.

High dose-rate (HDR) brachytherapy as monotherapy is an effective treatment for patients with low- and intermediate-risk prostate cancer and is increasingly being offered as a 2-fraction protocol. There is a lack of consensus on the optimal dosimetric planning parameters to use, or whether there is any benefit summating dosimetric parameters from more than one implant. Our goal is to determine planning parameters associated with disease control, toxicity and health-related quality of life (HRQOL). Data were collected on 83 patients with low- and intermediate-risk prostate cancer who received 2 fractions of 13.5 Gy HDR brachytherapy without androgen-deprivation therapy as part of a randomized phase II clinical trial. An in-house deformable, registration algorithm was used to co-register and dose-summate the plans from both implants for each patient. Acute and late GU and GI toxicities were measured using CTCAE 4.0 and HRQOL was measured in urinary, bowel, sexual and hormonal domains using the EPIC scores. Treatment efficacy was assessed through PSA measurement and imaging with or without biopsy where indicated. Covariates included baseline clinical factors, disease characteristics and treatment dosimetric parameters. Cox proportional hazards was performed to evaluate covariates impact on treatment toxicity and efficacy, and logistic regression analysis evaluated covariates impact on HRQOL. Among the 83 patients, median prostate volume was 46.7cm3. Median summated planning target volume receiving 100% prescription dose (PTV V100%) was 97.4%, median PTV V150% 42.4% and median PTV V200% 15.5%. Median highest dose to the 1cm3 rectum (D1cc) was 66.9% of the prescription dose and median rectum V80% was 0.008cm3. Median urethral D1cc was 99.0% of the prescription dose, median urethral Dmax 121.7% and median urethral D10% 116.2%. Grade ≥2 GI toxicity was uncommon (3.7% acute and 8.5% late), but grade ≥2 GU toxicity was reported in 73.2% (acute) and 46.3% (late) patients. Rectum D1cc and V80% were found to be significantly associated with grade 2 or higher acute GI toxicity, while use of a-blocker at baseline was associated with grade ≥2 acute GU toxicity. Similarly, use of a-blocker was associated with late grade ≥2 GU toxicity, but with no dosimetric associations. No other variables were associated with treatment-related toxicities. Only rectum D1cc was significantly associated with changes in bowel EPIC scores. Estimated 5-year biochemical disease-free survival was 93.9% and 5-year cumulative incidence of local failure was 3.8%. HDR monotherapy with 27 Gy delivered in 2 fractions in treatment of prostate cancer is well tolerated with high rates of disease control and minimal toxicity. Dose summation between 2 fractions of HDR brachytherapy is feasible, with rectal dose predicting acute GI toxicity. The lack of association between dose metrics and urinary toxicity raises the potential for further dose escalation.

Intelligent Interactive Deformable Image Registration for Online Adaptive Radiotherapy.

The goal of this study is to streamline the time-consuming contouring process in online adaptive radiotherapy (ART) by utilizing a deep learning-based interactive deformable image registration (DIR) algorithm. The objective is to minimize manual review and editing of automatically generated initial contours of organs-at-risk (OARs) and targets, thereby improving the efficiency and effectiveness of the treatment process. Our proposed method reforms the current DIR-based contour propagation method in clinical practice through the implementation of a deep learning-based interactive approach. The steps include: 1) generation of an initial deformable vector field (DVF) using a DL model, based on fixed and moving image pairs, resulting in the initial contours of OARs and targets; 2) clinician review/edit one the OAR/target contours as needed; 3) updated contour is sent to DL model to update the DVF and the remaining OARs/targets contours. Repeat this process until satisfactory contour qualities are achieved. We used the Open Access Series of Imaging Studies (OASIS) as the testbed, including 394 (train) and 20 (test) brain T1-weighted MRI scans, each containing 35 annotated organs. The U-Net architecture was employed to update the DVF from fixed/moving images, initial contours, and updated contours. We compared our approach to traditional manual editing without interaction and quantified the effort reduction using the added path length (APL) metric which is supposed to be proportional to the absolute time spent on the contour editing. We conducted paired t-test to show the significance. For comparison purpose, we assumed the clinicians edit the contours with the largest APL, i.e., the contours that require the most editing efforts. The editing effort, as measured by APL, was reduced by 18.5% to 25.4% with a mean of 23.3%, median of 23.6%, and standard deviation of 1.9%. The significance of the results was confirmed with a p-value of 1.47e-24. Our study demonstrates a significant reduction in editing effort, as measured by APL, compared to traditional manual contour editing. These results demonstrate the potential of our deep learning-based interactive approach to improve the efficiency and accuracy of the contouring process in clinical practice.

Use of Focal Radiotherapy Boost for Prostate Cancer and Perceived Barriers toward its Implementation: A Survey.

In a recent phase III randomized control trial (FLAME), delivering a focal radiotherapy (RT) boost to tumors visible on MRI was shown to improve outcomes for prostate cancer patients without increasing toxicity. The aim of this study was to assess how widely this technique is being applied in current practices worldwide as well as physicians' perceived barriers toward its implementation. An online survey assessing the use of intraprostatic focal boost was conducted in December 2022 and February 2023. The survey link was distributed to radiation oncologists worldwide via email list, group text platform, and social media. Survey questions included how many prostate cancer cases participants treat in a typical month; how often they use focal boost, if at all; the degree to which their practice is genitourinary (GU)-subspecialized; main barriers to implementing focal boost more often in their practice; and demographic information. Subgroup analyses were also conducted for participants from high-income or low-to-middle-income countries, as defined by the World Bank. The survey initially collected 205 responses from various countries over a two-week period in December 2022. The survey was then reopened for one week in February 2023 to allow for more participation, leading to a total of 263 responses. The highest-represented countries were the United States (42%), Mexico (13%), and the United Kingdom (8%). The majority of respondents worked at an academic medical center (52%) and considered their practice to be at least partially GU-subspecialized (74%). 57% of participants overall reported not routinely using intraprostatic focal boost. Even among complete subspecialists, a substantial proportion (39%) do not routinely use focal boost. Less than half of participants in both high-income and low-to-middle-income countries were shown to routinely use focal boost. Perceived barriers to implementation are shown in Table 1. Despite the promising level 1 results of the FLAME trial, many radiation oncologists worldwide are not routinely offering focal RT boost. Adoption of this technique might be accelerated by increased access to high-quality MRI, better registration algorithms of MRI to CT simulation images, physician education on benefit-to-harm ratio, automated planning algorithms, and physician training on contouring prostate lesions on MRI.

Robustness Analysis of Reference Surfaces for Surface Guided Radiation Therapy of the Breast

Surface guided radiation therapy (SGRT) implements an optical imaging system in radiation therapy for positioning and motion management. This system projects visible light onto a patient and the reflected light is used to generate 3D positional information so that clinicians can accurately reproduce body positions. Patient setup shifts are calculated with six degrees of freedom by a registration algorithm comparing a reference surface (RS) of the patient to a live surface map of the patient on treatment day. SGRT has been an effective tool in daily localization for the treatment of breast cancer patients. It is common for patients to have multiple RS throughout the course of their treatment to account for anatomical variation between fractions. We sought to evaluate the robustness of reference surfaces and vendor specific algorithms used for SGRT. At our institution, positional shift data for five patients treated for right-sided breast cancer were retrospectively analyzed. SGRT performance was compared between RS using bilateral breasts or a single ipsilateral breast. Shift parameters were calculated over the entire treatment course for all patients with a vendor supplied software tool that offers rigid and deformable registration algorithms. The deformable algorithm was used for treatment setups, with the treatment RS encompassing both breasts plus a margin. Two robustness tests were carried out: 1) a trimmed down RS encompassing just the ipsilateral breast and 2) a comparison of deformable vs rigid registration of the clinically used RS. After obtaining translational and angular shift data, the absolute mean differences between shifts were calculated to compare differences between RS size and algorithm performance. On average, 1.4 new RS were created per patient guided by weekly radiographic imaging to adjust for anatomical changes. The absolute value of the average of the discrepancies between shifts using the clinical RS subtracted from the trimmed external (89 fractions) were <1mm and 1° and the maximum differences were: Lateral: 2.6mm, Longitudinal: 1.4mm, Vertical: 1.1mm, Yaw: 1.1°, Roll: 1.5°, Pitch: 1.7°. Discrepancies between tracking algorithms (83 fractions) were <1.5mm and 1° and the maximum differences were: Lateral: 3.4mm, Longitudinal: 3.5mm, Vertical: 2.0mm, Yaw: 2.4°, Roll: 2.7°, Pitch: 1.9°. Clinically negligible mean discrepancies were observed for both robustness tests showing that neither the reference surface size nor the algorithms investigated caused systematic variations in the shifts for this group of patients. Maximum discrepancies of up to 3 mm and 3° were found between the algorithms, which indicate some variation, but within clinical tolerance. Overall, different selection of reference surfaces and algorithms had a minor effect on clinical shifts for SGRT of the breast.

Adaptive Auto-Segmentation for MRI-Guided Online Adaptive Radiotherapy of Cervical Carcinoma

Accurate and efficient delineation of organs and targets on session images is critical in MRI-guided online adaptive radiotherapy (MRgOART). This study proposes a registration-guided deep learning image segmentation framework to assist online delineation of cervical carcinoma. A total of 300 T2-weighted MR images were acquired for patients with cervical carcinoma treated by a 1.5T Unity MR-Linac. The CTV, bladder, rectum, pelvic bone and femoral joints were delineated on each MRI by the same radiation oncologist. To overcome these obstacles to online MRI segmentation, we propose a registration-guided DL (RgDL) segmentation framework that integrates image registration algorithms and DL segmentation models. Firstly, the DL segmentation model was trained using nnU-net. Then, for each treatment fraction, the deformable image registration (DIR) algorithm generates initial contours from previous treatment fraction, which were used as guidance by DL model to obtain the accurate current segmentation. The segmentation accuracy of alone DIR, DL and RgDL were evaluated by dice similarity coefficients (DSC) and other distance-based metrics. Compared to the baseline approaches using the DIR and the DL alone, RgDL achieved a DSC of 91.12% on CTV, higher than DIR and DL alone by 15.54% and 10.13%. The DSC of RgDL were improved to 95.58%, 93.65%, 87.8% and 94.84% for bladder, pelvic bone, rectum and femoral joints, higher than DIR and DL alone by 9.61% on average. The proposed adaptive auto-segmentation method can achieve accurate and efficient segmentation and potentially overcome these obstacles to MRgOART.

Deep Learning-Based Multi-Modality Segmentation of Primary Gross Tumor Volume in CT and MRI for Nasopharyngeal Carcinoma

The delineation of primary gross tumor volume (GTV) of nasopharyngeal carcinoma (NPC) is an essential step for radiotherapy planning. In clinical practice, radiation oncologists manually delineate the GTV in planning CT with the help of diagnostic MRI. This is because NPC tumors are closely adjacent to many important anatomic structures, and CT and MRI provide complementary strength to accurately determine the tumor extension boundary. Manual delineation is time-consuming with the potential registration errors between MRI and CT decreasing the delineation accuracy. In this study, we propose a fully automated GTV segmentation method based on CT and MRI by first aligning MRI to CT, and then, segmenting the GTV using a multi-modality deep learning model. We collected 104 nasopharyngeal carcinoma patients with both planning CT and diagnostic MRI scans (T1 & T2 phases). An experienced radiation oncologists manually delineated the GTV, which was further examined by another senior radiation oncologist. Then, a coarse to fine cross-modality registration from MRI to CT was conducted as follows: (1) A rigid transformation was performed on MRI to roughly align MRI to CT with similar anatomic position. (2) Then, the region of interest (RoI) on both CT and rigid-transformed MRI were cropped. (3) A leading cross-modality deformable registration algorithm, named DEEDS, was applied on the cropped MRI and CT RoIs to find an accurate local alignment. Next, using CT and registered MRI as the combined input, a multi-modality deep segmentation network based on nnUNet was trained to generate the GTV prediction. 20% patients were randomly selected as the unseen testing set to quantitatively evaluate the performance. The quantitative NPC GTV segmentation performance is summarized in Table 1. The deep segmentation model using CT alone achieved reasonable high performance with 76.6% Dice score and 1.34mm average surface distance (ASD). When both CT and registered MRI were used, the segmentation model further improved the performance by 0.9% Dice score increase and 11% relative ASD error reduction, demonstrating the complementary strength of CT and MRI in determining NPC GTV. Notably, the achieved 77.5% Dice score and 1.19mm ASD by the multimodality model is among the top performing results reported in recent automatic NPC GTV segmentation using either CT or MRI modality. We developed a fully automated multi-modal deep-learning model for NPC GTV segmentation. The developed model can segment the NPC GTV in high accuracy. With further optimization and validation, this automated model has potential to standardize the NPC GTV segmentation and significantly decrease the workload of radiation oncologists in clinical practice.