Deep Learning-Based Fine-Tuning Approach of Coarse Registration for Ear-Nose-Throat (ENT) Surgical Navigation Systems.

Accurate registration between medical images and patient anatomy is crucial for surgical navigation systems in minimally invasive surgeries. This study introduces a novel deep learning-based refinement step to enhance the accuracy of surface registration without disrupting established workflows. The proposed method integrates a machine learning model between conventional coarse registration and ICP fine registration. A deep-learning model was trained using simulated anatomical landmarks with introduced localization errors. The model architecture features global feature-based learning, an iterative prediction structure, and independent processing of rotational and translational components. Validation with silicon-masked head phantoms and CT imaging compared the proposed method to both conventional registration and a recent deep-learning approach. The results demonstrated significant improvements in target registration error (TRE) across different facial regions and depths. The average TRE for the proposed method (1.58 ± 0.52 mm) was significantly lower than that of the conventional (2.37 ± 1.14 mm) and previous deep-learning (2.29 ± 0.95 mm) approaches (p < 0.01). The method showed a consistent performance across various facial regions and enhanced registration accuracy for deeper areas. This advancement could significantly enhance precision and safety in minimally invasive surgical procedures.

Characterizing Low-cost Registration for Photographic Images to Computed Tomography.

Mapping information from photographic images to volumetric medical imaging scans is essential for linking spaces with physical environments, such as in image-guided surgery. Current methods of accurate photographic image to computed tomography (CT) image mapping can be computationally intensive and/or require specialized hardware. For general purpose 3-D mapping of bulk specimens in histological processing, a cost-effective solution is necessary. Here, we compare the integration of a commercial 3-D camera and cell phone imaging with a surface registration pipeline. Using surgical implants and chuck-eye steak as phantom tests, we obtain 3-D CT reconstruction and sets of photographic images from two sources: Canfield Imaging's H1 camera and an iPhone 14 Pro. We perform surface reconstruction from the photographic images using commercial tools and open-source code for Neural Radiance Fields (NeRF) respectively. We complete surface registration of the reconstructed surfaces with the iterative closest point (ICP) method. Manually placed landmarks were identified at three locations on each of the surfaces. Registration of the Canfield surfaces for three objects yields landmark distance errors of 1.747, 3.932, and 1.692 mm, while registration of the respective iPhone camera surfaces yields errors of 1.222, 2.061, and 5.155 mm. Photographic imaging of an organ sample prior to tissue sectioning provides a low-cost alternative to establish correspondence between histological samples and 3-D anatomical samples.

Longitudinally consistent registration and parcellation of cortical surfaces using semi-supervised learning.

Temporally consistent and accurate registration and parcellation of longitudinal cortical surfaces is of great importance in studying longitudinal morphological and functional changes of human brains. However, most existing methods are developed for registration or parcellation of a single cortical surface. When applying to longitudinal studies, these methods independently register/parcellate each surface from longitudinal scans, thus often generating longitudinally inconsistent and inaccurate results, especially in small or ambiguous cortical regions. Essentially, longitudinal cortical surface registration and parcellation are highly correlated tasks with inherently shared constraints on both spatial and temporal feature representations, which are unfortunately ignored in existing methods. To this end, we unprecedentedly propose a novel semi-supervised learning framework to exploit these inherent relationships from limited labeled data and extensive unlabeled data for more robust and consistent registration and parcellation of longitudinal cortical surfaces. Our method utilizes the spherical topology characteristic of cortical surfaces. It employs a spherical network to function as an encoder, which extracts high-level cortical features. Subsequently, we build two specialized decoders dedicated to the tasks of registration and parcellation, respectively. To extract more meaningful spatial features, we design a novel parcellation map similarity loss to utilize the relationship between registration and parcellation tasks, i.e., the parcellation map warped by the deformation field in registration should match the atlas parcellation map, thereby providing extra supervision for the registration task and augmented data for parcellation task by warping the atlas parcellation map to unlabeled surfaces. To enable temporally more consistent feature representation, we additionally enforce longitudinal consistency among longitudinal surfaces after registering them together using their concatenated features. Experiments on two longitudinal datasets of infants and adults have shown that our method achieves significant improvements on both registration/parcellation accuracy and longitudinal consistency compared to existing methods, especially in small and challenging cortical regions.

SUGAR: Spherical ultrafast graph attention framework for cortical surface registration.

Cortical surface registration plays a crucial role in aligning cortical functional and anatomical features across individuals. However, conventional registration algorithms are computationally inefficient. Recently, learning-based registration algorithms have emerged as a promising solution, significantly improving processing efficiency. Nonetheless, there remains a gap in the development of a learning-based method that exceeds the state-of-the-art conventional methods simultaneously in computational efficiency, registration accuracy, and distortion control, despite the theoretically greater representational capabilities of deep learning approaches. To address the challenge, we present SUGAR, a unified unsupervised deep-learning framework for both rigid and non-rigid registration. SUGAR incorporates a U-Net-based spherical graph attention network and leverages the Euler angle representation for deformation. In addition to the similarity loss, we introduce fold and multiple distortion losses to preserve topology and minimize various types of distortions. Furthermore, we propose a data augmentation strategy specifically tailored for spherical surface registration to enhance the registration performance. Through extensive evaluation involving over 10,000 scans from 7 diverse datasets, we showed that our framework exhibits comparable or superior registration performance in accuracy, distortion, and test-retest reliability compared to conventional and learning-based methods. Additionally, SUGAR achieves remarkable sub-second processing times, offering a notable speed-up of approximately 12,000 times in registering 9,000 subjects from the UK Biobank dataset in just 32 min. This combination of high registration performance and accelerated processing time may greatly benefit large-scale neuroimaging studies.

Automatic landmark detection and registration of brain cortical surfaces via quasi-conformal geometry and convolutional neural networks.

In medical imaging, surface registration is extensively used for performing systematic comparisons between anatomical structures, with a prime example being the highly convoluted brain cortical surfaces. To obtain a meaningful registration, a common approach is to identify prominent features on the surfaces and establish a low-distortion mapping between them with the feature correspondence encoded as landmark constraints. Prior registration works have primarily focused on using manually labeled landmarks and solving highly nonlinear optimization problems, which are time-consuming and hence hinder practical applications. In this work, we propose a novel framework for the automatic landmark detection and registration of brain cortical surfaces using quasi-conformal geometry and convolutional neural networks. We first develop a landmark detection network (LD-Net) that allows for the automatic extraction of landmark curves given two prescribed starting and ending points based on the surface geometry. We then utilize the detected landmarks and quasi-conformal theory for achieving the surface registration. Specifically, we develop a coefficient prediction network (CP-Net) for predicting the Beltrami coefficients associated with the desired landmark-based registration and a mapping network called the disk Beltrami solver network (DBS-Net) for generating quasi-conformal mappings from the predicted Beltrami coefficients, with the bijectivity guaranteed by quasi-conformal theory. Experimental results are presented to demonstrate the effectiveness of our proposed framework. Altogether, our work paves a new way for surface-based morphometry and medical shape analysis.

"Image to patient" equal-resolution surface registration supported by a surface scanner: analysis of algorithm efficiency for computer-aided surgery.

PURPOSE: The "image to patient" registration procedure is crucial for the accuracy of surgical instrument tracking relative to the medical image while computer-aided surgery. The main aim of this work was to create an equal-resolution surface registration algorithm (ERSR) and analyze its efficiency. METHODS: The ERSR algorithm provides two datasets with equal, high resolution and approximately corresponding points. The registered sets are obtained by projection of a user-designed rectangle(s)-shaped uniform clouds of points on DICOM and surface scanner datasets. The tests of the algorithm were performed on a phantom with titanium microscrews. We analyzed the influence of DICOM resolution on the effect of the ERSR algorithm and compared the ERSR to standard paired-points landmark transform registration. The methods of analysis were Target Registration Error, distance maps, and their histogram evaluation. RESULTS: The mean TRE in case of ERSR equaled 0.8 ± 0.3 mm (resolution A), 0.8 ± 0.5 mm (resolution B), and 1.0 ± 0.7 mm (resolution C). The mean values were at least 0.4 mm lower than in the case of landmark transform registration. The distance maps between the model achieved from the scanner and the CT-based model were analyzed by histogram. The frequency of the first bin in a histogram of the distance map for ERSR was about 0.6 for all three resolutions of DICOM dataset and three times higher than in the case of landmark transform registration. The results were statistically analyzed using the Wilcoxon signed-rank test (alpha = 0.05). CONCLUSION: The tests proved a statistically significant higher efficiency of equal resolution surface registration related to the landmark transform algorithm. It was proven that the lower resolution of the CT DICOM dataset did not degrade the efficiency of the ERSR algorithm. We observed a significantly lower response to decreased resolution than in the case of paired-points landmark transform registration.

The threshold of an intra oral scanner to measure lesion depth on natural unpolished teeth.

OBJECTIVES: To investigate the threshold and accuracy of intraoral scanning in measuring freeform human enamel surfaces. METHODS: Software softgauges, ranging between 20 and 160 µm depth, were used to compare four workflow analysis techniques to measure step height on a freeform surface; with or without reference areas and in combination with surface-subtraction to establish which combination produced the most accurate outcome. Having established the optimum combination, 1.5 mm diameter, individual depths ranging from 11 to 81 µm were created separately on 14 unpolished human enamel samples and then scanned with gold standard laboratory optical profilometry (NCLP, TaiCaan Technologies™, XYRIS2000CL, UK) and a clinical intraoral scanner (TrueDefinition™, Midmark Corp., USA). The sequence of surface registration and subtraction determined from the softgauges was used to measure step height on natural human enamel surfaces. Step heights (µm) were compared using two-way ANOVA with post-hoc Bonferroni (p < 0.05) and Bland-Altman analyses. RESULTS: Software differences were significantly reduced from - 29.7 to - 32.5% without, to - 2.4 to - 3.6% with reference areas (p < 0.0001) and the addition of surface-subtraction after registration reduced this further to 0.0 to - 0.3% (p < 0.0001). The intraoral scanner had a depth discrimination threshold of 73 µm on unpolished natural enamel and significant differences (p < 0.05) were observed compared to NCLP below this level. SIGNIFICANCE: The workflow of combining surface-registration and subtraction of surface profiles taken from intraoral scans of freeform unpolished enamel enabled confident measurement of step height above 73 µm. The limits of the scanner is related to data capture and these results provide opportunities for clinical measurement.

Automated machine learning (AutoML)-based surface registration methodology for image-guided surgical navigation system.

BACKGROUND: Although the surface registration technique has the advantage of being relatively safe and the operation time is short, it generally has the disadvantage of low accuracy. PURPOSE: This research proposes automated machine learning (AutoML)-based surface registration to improve the accuracy of image-guided surgical navigation systems. METHODS: The state-of-the-art surface registration concept is that first, using a neural network model, a new point-cloud that matches the facial information acquired by a passive probe of an optical tracking system (OTS) is extracted from the facial information obtained by computerized tomography. Target registration error (TRE) representing the accuracy of surface registration is then calculated by applying the iterative closest point (ICP) algorithm to the newly extracted point-cloud and OTS information. In this process, the hyperparameters used in the neural network model and ICP algorithm are automatically optimized using Bayesian optimization with expected improvement to yield improved registration accuracy. RESULTS: Using the proposed surface registration methodology, the average TRE for the targets located in the sinus space and nasal cavity of the soft phantoms is 0.939 ± 0.375 mm, which shows 57.8% improvement compared to the average TRE of 2.227 ± 0.193 mm calculated by the conventional surface registration method (p < 0.01). The performance of the proposed methodology is evaluated, and the average TREs computed by the proposed methodology and the conventional method are 0.767 ± 0.132 and 2.615 ± 0.378 mm, respectively. Additionally, for one healthy adult, the clinical applicability of the AutoML-based surface registration is also presented. CONCLUSION: Our findings showed that the registration accuracy could be improved while maintaining the advantages of the surface registration technique.

Fusion of multimodality image and point cloud for spatial surface registration for knee arthroplasty.

BACKGROUND: Image-guided computer-aided navigation system is an indispensable part of computer assisted orthopaedic surgery. However, the location and number of fiducial markers, the time required to localise fiducial markers in existing systems affect their effectiveness. METHOD: The study proposed that spatial surface registration between the point cloud on the surface of the fusion model based on preoperative knee MRI and CT images and the point cloud on the cartilage surface captured by intraoperative laser scanner could solve the above limitations. RESULTS: The experimental results show that the registration error of the method is less than 2 mm, but the total time from scanning the point cloud on patient's cartilage surface to registering it with the point cloud in preoperative image space is less than 2 min. CONCLUSION: The method achieves the registration accuracy similar to existing methods without selecting anatomical corresponding points, which is of great help to the clinic.

Increase in tibial internal rotation due to weight-bearing is a key feature to diagnose early-stage knee osteoarthritis: a study with upright computed tomography.

BACKGROUND: The classification of knee osteoarthritis is an essential clinical issue, particularly in terms of diagnosing early knee osteoarthritis. However, the evaluation of three-dimensional limb alignment on two-dimensional radiographs is limited. This study evaluated the three-dimensional changes induced by weight-bearing in the alignments of lower limbs at various stages of knee osteoarthritis. METHODS: Forty five knees of 25 patients (69.9 ± 8.9 years) with knee OA were examined in the study. CT images of the entire leg were obtained in the supine and standing positions using conventional CT and 320-row detector upright CT, respectively. Next, the differences in the three-dimensional alignment of the entire leg in the supine and standing positions were obtained using 3D-3D surface registration technique, and those were compared for each Kellgren-Lawrence grade. RESULTS: Greater flexion, adduction, and tibial internal rotation were observed in the standing position, as opposed to the supine position. Kellgren-Lawrence grades 1 and 4 showed significant differences in flexion, adduction, and tibial internal rotation between two postures. Grades 2 and 4 showed significant differences in adduction, while grades 1 and 2, and 1 and 3 showed significant differences in tibial internal rotation between standing and supine positions. CONCLUSIONS: Weight-bearing makes greater the three-dimensional deformities in knees with osteoarthritis. Particularly, greater tibial internal rotation was observed in patients with grades 2 and 3 compared to those with grade 1. The greater tibial internal rotation due to weight-bearing is a key pathologic feature to detect early osteoarthritic change in knees undergoing osteoarthritis.

Accuracy evaluation of surface registration algorithm using normal distribution transform in stereotactic body radiotherapy/radiosurgery: A phantom study.

PURPOSE: To evaluate a feasibility of normal distribution transform (NDT) algorithm compared with the iterative closest point (ICP) method as a useful surface registration in stereotactic body radiotherapy (SBRT)/stereotactic radiosurgery (SRS). METHODS: Point cloud images using the 3D triangulation technology were obtained from a depth camera-based optical imaging (OSI) system equipped in a radiosurgery room. Two surface registration algorithms, NDT and ICP, were used to measure and compare the discrepancy values between the reference and the current surfaces during the positioning of the patient. The performance evaluation was investigated by calculating the registration error and root-mean-square (RMS) values for the surface model, reposition, and target accuracy, which were analyzed statistically using a paired t-test. RESULTS: For surface model accuracy, the average of the registration error and RMS values were measured as 3.56 ± 2.20 mm and 6.98 ± 1.89 mm for ICP method, and 1.76 ± 1.32 mm and 3.58 ± 1.30 mm for NDT method (p < 0.05). For reposition accuracy, the average registration error and RMS values were calculated as 1.41 ± 0.98 mm and 2.53 ± 1.64 mm using ICP method, and 0.92 ± 0.61 mm and 1.75 ± 0.80 mm using NDT method (p = 0.005). The overall target accuracy using the NDT method reduced the average of the reposition error and overall RMS value by 0.71 and 1.32 mm, respectively, compared to the ICP method (p = 0.03). CONCLUSIONS: We found that the surface registration algorithm based on NDT method provides more reliable accuracy in the values of surface model, reposition, and target accuracies than the classic ICP method. The NDT method in OSI systems offers reasonable accuracy in SBRT/SRS.

Comparison of radiation exposure and surgery time between an intraoperative CT with automatic surface registration and a preoperative CT with manual surface registration in navigated spinal surgeries.

PURPOSE: This retrospective matched case-control study was conducted to compare two CT based surgery techniques for navigated screw placement in spinal surgery, whether a reduction of radiation exposure and surgery time could be achieved. METHODS: We matched cases treated with an intraoperative CT (iCT), regarding the type and number of implants, with cases treated with a preoperative CT (pCT) of one main surgeon. Outcome measures were radiation exposure due to intraoperative control x-rays, radiation exposure due to CT images, and the duration of surgery. RESULTS: The required radiation exposure could be significantly reduced in the iCT group. For the intraoperative control X-rays by 69% (median (MED) 88.50/standard deviation (SD) 107.84 and MED 286.00/SD 485.04 for iCT and pCT respectively-in Gycm2; p < 0.001) and for the CT examinations by 25% (MED 317.00/SD 158.62 and MED 424.50/SD 225.04 for iCT and pCT respectively-in mGycm; p < 0.001) with no significant change in surgery time. The correlation between the number of segments fused and the necessary surgery time decreased significantly for the iCT group (Pearson product-moment-correlation: r = 0.569 and r = 0.804 for iCT and pCT respectively; p < 0.05). CONCLUSION: The results show that spinal navigation using an intraoperative CT with automatic registration compared to a preoperative CT and intraoperative manual surface registration, allows a significant reduction of radiation exposure, without prolonged surgery time. A significant benefit regarding cut-to-suture-time can be gained with surgeries of a larger scale.

PDE-constrained shape registration to characterize biological growth and morphogenesis from imaging data.

We propose a PDE-constrained shape registration algorithm that captures the deformation and growth of biological tissue from imaging data. Shape registration is the process of evaluating optimum alignment between pairs of geometries through a spatial transformation function. We start from our previously reported work, which uses 3D tensor product B-spline basis functions to interpolate 3D space. Here, the movement of the B-spline control points, composed with an implicit function describing the shape of the tissue, yields the total deformation gradient field. The deformation gradient is then split into growth and elastic contributions. The growth tensor captures addition of mass, i.e. growth, and evolves according to a constitutive equation which is usually a function of the elastic deformation. Stress is generated in the material due to the elastic component of the deformation alone. The result of the registration is obtained by minimizing a total energy functional which includes: a distance measure reflecting similarity between the shapes, and the total elastic energy accounting for the growth of the tissue. We apply the proposed shape registration framework to study zebrafish embryo epiboly process and tissue expansion during skin reconstruction surgery. We anticipate that our PDE-constrained shape registration method will improve our understanding of biological and medical problems in which tissues undergo extreme deformations over time.

A comparison of voxel- and surface-based cone-beam computed tomography mandibular superimposition in adult orthodontic patients.

OBJECTIVE: To evaluate the accuracy, reliability, and efficiency of voxel- and surface-based registrations for cone-beam computed tomography (CBCT) mandibular superimposition in adult orthodontic patients. METHODS: Pre- and post-orthodontic treatment CBCT scans of 27 adult patients were obtained. Voxel- and surface-based CBCT mandibular superimpositions were performed using the mandibular basal bone as a reference. The accuracy of the two methods was evaluated using the absolute mean distance measured. The time that was required to perform the measurements using these methods was also compared. Statistical differences were determined using paired t-tests, and inter-observer reliability was assessed by intraclass correlation coefficients (ICCs). RESULTS: The absolute mean distance on seven mandible surface areas between voxel- and surface-based registrations was similar but not significantly different. ICC values of the surface-based registration were 0.918 to 0.990, which were slightly lower than those of voxel-based registration that ranged from 0.984 to 0.996. The time required for voxel-based registration and surface-based registration was 44.6 ± 2.5 s and 252.3 ± 7.1 s, respectively. CONCLUSIONS: Both methods are accurate and reliable and not significantly different from each other. However, voxel-based registration is more efficient than surface-based registration for CBCT mandibular superimposition.

A Deep Network for Joint Registration and Parcellation of Cortical Surfaces.

Cortical surface registration and parcellation are two essential steps in neuroimaging analysis. Conventionally, they are performed independently as two tasks, ignoring the inherent connections of these two closely-related tasks. Essentially, both tasks rely on meaningful cortical feature representations, so they can be jointly optimized by learning shared useful cortical features. To this end, we propose a deep learning framework for joint cortical surface registration and parcellation. Specifically, our approach leverages the spherical topology of cortical surfaces and uses a spherical network as the shared encoder to first learn shared features for both tasks. Then we train two task-specific decoders for registration and parcellation, respectively. We further exploit the more explicit connection between them by incorporating the novel parcellation map similarity loss to enforce the boundary consistency of regions, thereby providing extra supervision for the registration task. Conversely, parcellation network training also benefits from the registration, which provides a large amount of augmented data by warping one surface with manual parcellation map to another surface, especially when only few manually-labeled surfaces are available. Experiments on a dataset with more than 600 cortical surfaces show that our approach achieves large improvements on both parcellation and registration accuracy (over separately trained networks) and enables training high-quality parcellation and registration models using much fewer labeled data.

Learning 4D Infant Cortical Surface Atlas with Unsupervised Spherical Networks.

Spatiotemporal (4D) cortical surface atlas during infancy plays an important role for surface-based visualization, normalization and analysis of the dynamic early brain development. Conventional atlas construction methods typically rely on classical group-wise registration on sub-populations and ignore longitudinal constraints, thus having three main issues: 1) constructing templates at discrete time points; 2) resulting in longitudinal inconsistency among different age's atlases; and 3) taking extremely long runtime. To address these issues, in this paper, we propose a fast unsupervised learning-based surface atlas construction framework incorporating longitudinal constraints to enforce the within-subject temporal correspondence in the atlas space. To well handle the difficulties of learning large deformations, we propose a multi-level multimodal spherical registration network to perform cortical surface registration in a coarse-to-fine manner. Thus, only small deformations need to be estimated at each resolution level using the registration network, which further improves registration accuracy and atlas quality. Our constructed 4D infant cortical surface atlas based on 625 longitudinal scans from 291 infants is temporally continuous, in contrast to the state-of-the-art UNC 4D Infant Surface Atlas, which only provides the atlases at a few discrete sparse time points. By evaluating the intra- and inter-subject spatial normalization accuracy after alignment onto the atlas, our atlas demonstrates more detailed and fine-grained cortical patterns, thus leading to higher accuracy in surface registration.

CIVET-Macaque: An automated pipeline for MRI-based cortical surface generation and cortical thickness in macaques.

The MNI CIVET pipeline for automated extraction of cortical surfaces and evaluation of cortical thickness from in-vivo human MRI has been extended for processing macaque brains. Processing is performed based on the NIMH Macaque Template (NMT), as the reference template, with the anatomical parcellation of the surface following the D99 and CHARM atlases. The modifications needed to adapt CIVET to the macaque brain are detailed. Results have been obtained using CIVET-macaque to process the anatomical scans of the 31 macaques used to generate the NMT and another 95 macaques from the PRIME-DE initiative. It is anticipated that the open usage of CIVET-macaque will promote collaborative efforts in data collection and processing, sharing, and automated analyses from which the non-human primate brain imaging field will advance.

Validation of a commercial deformable image registration for surface-guided radiotherapy using an ad hoc-developed deformable phantom.

PURPOSE: The use of optical surface systems (OSSs) for patient setup verification in external radiation therapy is increasing. To manage potential deformations in a patient's anatomy, a novel deformable image registration (DIR) tool has been applied in a commercial OSS. In this study we investigate the accuracy of the DIR as compared to rigid image registration (RR). METHODS AND MATERIALS: The positioning accuracy of the DIR and RR implemented in the OSS was investigated using an ad hoc-developed anthropomorphic deformable phantom, named Mary. The phantom consists of 33 slices of expanded polystyrene slabs shaped thus to simulate part of a female body. Anatomical details, simulating the ribs and spinal cord, together with 10 inner targets at different depths are included in thorax and abdominal parts. Mary is capable of realistic body movements and deformations, such as head and arm rotations, body torsion and moderate breast/abdomen swelling. The accuracy of DIR and RR was investigated for four internal targets after deliberately deforming the phantom nine times. Breast and abdomen enlargements and torsions around x, y, and z axes were applied. For reference purposes, rigid displacements (where Mary's anatomy was kept intact) were included. The phantom was positioned on the linac couch under the OSS guidance and for each target and displacement a CBCT was acquired. The accuracy of DIR and RR was assessed evaluating the difference in means of absolute values between CBCT and the OSS registration parameters (lateral, longitudinal, vertical, rot, pitch, and roll), using both a reference surface extracted from CT (CTr) or acquired with the OSS (OSSr). A comparison of the four different combinations, DIR + OSSr, DIR + CTr, RR + OSSr, and RR + CTr, was carried out to evaluate the position accuracy for the various combinations. Finally, the positioning accuracy of the different target positions using only OSSr was investigated for the DIR. A paired sample Wilcoxon signed-rank test (P < 0.05) and a two-tailed Mann-Whitney test (P < 0.05) were carried out. RESULTS: The DIR in combination with OSSr showed significantly (P < 0.05) improved positioning accuracy in the lateral and longitudinal directions and in pitch, compared to RR, when deformations were applied to Mary. The positioning accuracy improved from 1.9 ± 1.5 mm, 1.1 ± 0.8 mm to 1.1 ± 1.2 mm, 0.6 ± 0.5 mm in lateral and longitudinal directions, respectively, and from 0.8 ± 0.6° to 0.4 ± 0.4° in pitch, using DIR compared to RR. Both the DIR and RR showed a similar positioning accuracy when rigid displacements of Mary were applied. For DIR, the OSSr generally showed improved calculation accuracy compared to CTr. Independent of the reference image used, the target position influenced the registration accuracy, and hence, one target could not be evaluated using RR due to its inability to calculate the correct position. CONCLUSIONS: Improved positioning accuracy was observed for DIR with respect to RR when deformations of Mary's anatomy were applied. For both DIR and RR, improved positioning accuracy was observed using OSSr as compared to CTr. The position of the target inside the phantom influenced the positioning accuracy for DIR.

Cortical surface registration using unsupervised learning.

Non-rigid cortical registration is an important and challenging task due to the geometric complexity of the human cortex and the high degree of inter-subject variability. A conventional solution is to use a spherical representation of surface properties and perform registration by aligning cortical folding patterns in that space. This strategy produces accurate spatial alignment, but often requires high computational cost. Recently, convolutional neural networks (CNNs) have demonstrated the potential to dramatically speed up volumetric registration. However, due to distortions introduced by projecting a sphere to a 2D plane, a direct application of recent learning-based methods to surfaces yields poor results. In this study, we present SphereMorph, a diffeomorphic registration framework for cortical surfaces using deep networks that addresses these issues. SphereMorph uses a UNet-style network associated with a spherical kernel to learn the displacement field and warps the sphere using a modified spatial transformer layer. We propose a resampling weight in computing the data fitting loss to account for distortions introduced by polar projection, and demonstrate the performance of our proposed method on two tasks, including cortical parcellation and group-wise functional area alignment. The experiments show that the proposed SphereMorph is capable of modeling the geometric registration problem in a CNN framework and demonstrate superior registration accuracy and computational efficiency. The source code of SphereMorph will be released to the public upon acceptance of this manuscript at https://github.com/voxelmorph/spheremorph.

Characterizing a deformable registration algorithm for surface-guided breast radiotherapy.

PURPOSE: Surface-guided radiation therapy (SGRT) is a nonionizing imaging approach for patient setup guidance, intra-fraction monitoring, and automated breath-hold gating of radiation treatments. SGRT employs the premise that the external patient surface correlates to the internal anatomy, to infer the treatment isocenter position at time of treatment delivery. Deformations and posture variations are known to impact the correlation between external and internal anatomy. However, the degree, magnitude, and algorithm dependence of this impact are not intuitive and currently no methods exist to assess this relationship. The primary aim of this work was to develop a framework to investigate and understand how a commercial optical surface imaging system (C-RAD, Uppsala, Sweden), which uses a nonrigid registration algorithm, handles rotations and surface deformations. METHODS: A workflow consisting of a female torso phantom and software-introduced transformations to the corresponding digital reference surface was developed. To benchmark and validate the approach, known rigid translations and rotations were first applied. Relevant breast radiotherapy deformations related to breast size, hunching/arching back, distended/deflated abdomen, and an irregular surface to mimic a cover sheet over the lower part of the torso were investigated. The difference between rigid and deformed surfaces was evaluated as a function of isocenter location. RESULTS: For all introduced rigid body transformations, C-RAD computed isocenter shifts were determined within 1 mm and 1˚. Additional translational shifts to correct for rotations as a function of isocenter location were determined with the same accuracy. For yaw setup errors, the difference in shift corrections between a plan with an isocenter placed in the center of the breast (BrstIso) and one located 12 cm superiorly (SCFIso) was 2.3 mm/1˚ in lateral direction. Pitch setup errors resulted in a difference of 2.1 mm/1˚ in vertical direction. For some of the deformation scenarios, much larger differences up to 16 mm and 7˚ in the calculated shifts between BrstIso and SCFIso were observed that could lead to large unintended gaps or overlap between adjacent matched fields if uncorrected. CONCLUSIONS: The methodology developed lends itself well for quality assurance (QA) of SGRT systems. The deformable C-RAD algorithm determined accurate shifts for rigid transformations, and this was independent of isocenter location. For surface deformations, the position of the isocenter had considerable impact on the registration result. It is recommended to avoid off-axis isocenters during treatment planning to optimally utilize the capabilities of the deformable image registration algorithm, especially when multiple isocenters are used with fields that share a field edge.