This paper aims at proposing a method to generate atlases of white matter fibers’ geometry that consider local orientation and curvature of fibers extracted from tractography data. Tractography was performed on diffusion magnetic resonance images from a set of healthy subjects and each tract was characterized voxel-wise by its curvature and Frenet–Serret frame, based on which similar tracts could be clustered separately for each voxel and each subject. Finally, the centroids of the clusters identified in all subjects were clustered to create the final atlas. The proposed clustering technique showed promising results in identifying voxel-wise distributions of curvature and orientation. Two tractography algorithms (one deterministic and one probabilistic) were tested for the present work, obtaining two different atlases. A high agreement between the two atlases was found in several brain regions. This suggests that more advanced tractography methods might only be required for some specific regions in the brain. In addition, the probabilistic approach resulted in the identification of a higher number of fiber orientations in various white matter areas, suggesting it to be more adequate for investigating complex fiber configurations in the proposed framework as compared to deterministic tractography.

Cortical bone plays a big role in the mechanical competence of bone. The analysis of cortical bone requires accurate segmentation methods. Level set methods are usually in the state-of-the-art for segmenting medical images. However, traditional implementations of this method are computationally expensive. This drawback was recently tackled through the so-called coherent propagation extension of the classical algorithm which has decreased computation times dramatically. In this study, we assess the potential of this technique for segmenting cortical bone in interactive time in 3D images acquired through High Resolution peripheral Quantitative Computed Tomography (HR-pQCT). The obtained segmentations are used to estimate cortical thickness and cortical porosity of the investigated images. Cortical thickness and Cortical porosity is computed using sphere fitting and mathematical morphological operations respectively. Qualitative comparison between the segmentations of our proposed algorithm and a previously published approach on six images volumes reveals superior smoothness properties of the level set approach. While the proposed method yields similar results to previous approaches in regions where the boundary between trabecular and cortical bone is well defined, it yields more stable segmentations in challenging regions. This results in more stable estimation of parameters of cortical bone. The proposed technique takes few seconds to compute, which makes it suitable for clinical settings.

The accuracy of the analyses for studying the three dimensional trabecular bone microstructure rely on the quality of the segmentation between trabecular bone and bone marrow. Such segmentation is challenging for images from computed tomography modalities that can be used in vivo due to their low contrast and resolution. For this purpose, we propose in this paper a granulometry-based segmentation method. In a first step, the trabecular thickness is estimated by using the granulometry in gray scale, which is generated by applying the opening morphological operation with ball-shaped structuring elements of different diameters. This process mimics the traditional sphere-fitting method used for estimating trabecular thickness in segmented images. The residual obtained after computing the granulometry is compared to the original gray scale value in order to obtain a measurement of how likely a voxel belongs to trabecular bone. A threshold is applied to obtain the final segmentation. Six histomorphometric parameters were computed on 14 segmented bone specimens imaged with cone-beam computed tomography (CBCT), considering micro-computed tomography (micro-CT) as the ground truth. Otsu’s thresholding and Automated Region Growing (ARG) segmentation methods were used for comparison. For three parameters (Tb.N, Tb.Th and BV/TV), the proposed segmentation algorithm yielded the highest correlations with micro-CT, while for the remaining three (Tb.Nd, Tb.Tm and Tb.Sp), its performance was comparable to ARG. The method also yielded the strongest average correlation (0.89). When Tb.Th was computed directly from the gray scale images, the correlation was superior to the binary-based methods. The results suggest that the proposed algorithm can be used for studying trabecular bone in vivo through CBCT.

Content-Based Medical Image Retrieval (CBMIR) is an important research field in the context of medical data management. In this paper we propose a novel CBMIR system for the automatic retrieval of radiographic images. Our approach employs a Convolutional Neural Network (CNN) to obtain high- level image representations that enable a coarse retrieval of images that are in correspondence to a query image. The retrieved set of images is refined via a non-parametric estimation of putative classes for the query image, which are used to filter out potential outliers in favour of more relevant images belonging to those classes. The refined set of images is finally re-ranked using Edge Histogram Descriptor, i.e. a low-level edge-based image descriptor that allows to capture finer similarities between the retrieved set of images and the query image. To improve the computational efficiency of the system, we employ dimensionality reduction via Principal Component Analysis (PCA). Experiments were carried out to evaluate the effectiveness of the proposed system on medical data from the “Image Retrieval in Medical Applications” (IRMA) benchmark database. The obtained results show the effectiveness of the proposed CBMIR system in the field of medical image retrieval.

Introduction: In the last few years, several models trying to calculate the biological brain age have been proposed based on structural magnetic resonance imaging scans (T1-weighted MRIs, T1w) using multivariate methods and machine learning. We developed and validated a convolutional neural network (CNN)-based biological brain age prediction model that uses one T1w MRI preprocessing step when applying the model to external datasets to simplify implementation and increase accessibility in research settings. Our model only requires rigid image registration to the MNI space, which is an advantage compared to previous methods that require more preprocessing steps, such as feature extraction. Methods: We used a multicohort dataset of cognitively healthy individuals (age range = 32.0–95.7 years) comprising 17,296 MRIs for training and evaluation. We compared our model using hold-out (CNN1) and cross-validation (CNN2–4) approaches. To verify generalisability, we used two external datasets with different populations and MRI scan characteristics to evaluate the model. To demonstrate its usability, we included the external dataset’s images in the cross-validation training (CNN3). To ensure that our model used only the brain signal on the image, we also predicted brain age using skull-stripped images (CNN4). Results: The trained models achieved a mean absolute error of 2.99, 2.67, 2.67, and 3.08 years for CNN1–4, respectively. The model’s performance in the external dataset was in the typical range of mean absolute error (MAE) found in the literature for testing sets. Adding the external dataset to the training set (CNN3), overall, MAE is unaffected, but individual cohort MAE improves (5.63–2.25 years). Salience maps of predictions reveal that periventricular, temporal, and insular regions are the most important for age prediction. Discussion: We provide indicators for using biological (predicted) brain age as a metric for age correction in neuroimaging studies as an alternative to the traditional chronological age. In conclusion, using different approaches, our CNN-based model showed good performance using one T1w brain MRI preprocessing step. The proposed CNN model is made publicly available for the research community to be easily implemented and used to study ageing and age-related disorders.

Rigid registration aims to determine the translations and rotations necessary to align features in a pair of images. While recent machine learning methods have become state-of-the-art for linear and deformable registration across subjects, they have demonstrated limitations when applied to longitudinal (within-subject) registration, where achieving precise alignment is critical. Building on an existing framework for anatomy-aware, acquisition-agnostic affine registration, we propose a model optimized for longitudinal, rigid brain registration. By training the model with synthetic within-subject pairs augmented with rigid and subtle nonlinear transforms, the model estimates more accurate rigid transforms than previous cross-subject networks and performs robustly on longitudinal registration pairs within and across magnetic resonance imaging (MRI) contrasts. 

Significant advances have been made toward building accurate automatic segmentation models for adult gliomas. However, the performance of these models often degrades when applied to pediatric glioma due to their imaging and clinical differences (domain shift). Obtaining sufficient annotated data for pediatric glioma is typically difficult because of its rare nature. Also, manual annotations are scarce and expensive. In this work, we propose Domain-Adapted nnU-Net (DA-nnUNet) to perform unsupervised domain adaptation from adult glioma (source domain) to pediatric glioma (target domain). Specifically, we add a domain classifier connected with a gradient reversal layer (GRL) to a backbone nnU-Net. Once the classifier reaches a very high accuracy, the GRL is activated with the goal of transferring domain-invariant features from the classifier to the segmentation model while preserving segmentation accuracy on the source domain. The accuracy of the classifier slowly degrades to chance levels. No annotations are used in the target domain. The method is compared to 8 different supervised models using BraTS-Adult glioma (N=1251) and BraTS-PED glioma data (N=99). The proposed method shows notable performance enhancements in the tumor core (TC) region compared to the model that only uses adult data: ~32% better Dice scores and ~20 better 95th percentile Hausdorff distances. Moreover, our unsupervised approach shows no statistically significant difference compared to the practical upper bound model using manual annotations from both datasets in TC region. The code is shared at https://github.com/Fjr9516/DA_nnUNet.

Alzheimer’s Disease (AD) and normal aging are both characterized by brain atrophy. The question of whether AD-related brain atrophy represents accelerated aging or a neurodegeneration process distinct from that in normal aging remains unresolved. Moreover, precisely disentangling AD-related brain atrophy from normal aging in a clinical context is complex. In this study, we propose a deformation-based morphometry framework to estimate normal aging and AD-specific atrophy patterns of subjects from morphological MRI scans. For this, we first leverage deep-learning-based methods to create age-dependent templates of cognitively normal (CN) subjects. These templates model the normal aging atrophy patterns in a CN population. Then, we use the learned diffeomorphic registration to estimate the one-year normal aging pattern at the voxel level. In the second step, we register the testing image to the 60-year-old CN template. Finally, normal aging and AD-specific scores are estimated by measuring the alignment of this registration with the one-year normal aging pattern. The methodology was developed and evaluated on the OASIS3 dataset with 1,014 T1-weighted MRI scans, which is a unique dataset focused on preclinical cohorts. Of these, 326 scans were from CNsubjects, and 688 scans were from individuals clinically diagnosed with AD at different stages of clinical severity defined by clinical dementia rating (CDR) scores. The results show that ventricles predominantly follow an accelerated normal aging pattern in subjects with AD. In turn, hippocampi and amygdala regions were affected by both normal aging and AD-specific factors. Interestingly, hippocampi and amygdala regions showed more of an accelerated normal aging pattern for subjects during the early clinical stages of the disease, while the AD-specific score increases in later clinical stages. Our code is freely available at https://github.com/Fjr9516/DBM_with_DL.

Predicting brain aging can help in the early detection and prognosis of neurodegenerative diseases. Longitudinal cohorts of healthy subjects scanned through magnetic resonance imaging (MRI) have been essential to understand the structural brain changes due to aging. However, these cohorts suffer from missing data due to logistic issues in the recruitment of subjects. This paper proposes a methodology for filling up missing data in longitudinal cohorts with anatomically plausible images that capture the subject-specific aging process. The proposed methodology is developed within the framework of diffeomorphic registration. First, two novel modules are introduced within Synthmorph, a fast, state-of-the-art deep learning-based diffeomorphic registration method, to simulate the aging process between the first and last available MRI scan for each subject in three-dimensional (3D). The use of image registration also makes the generated images plausible by construction. Second, we used six image similarity measurements to rearrange the generated images to the specific age range. Finally, we estimated the age of every generated image by using the assumption of linear brain decay in healthy subjects. The methodology was evaluated on 2662 T1-weighted MRI scans from 796 healthy participants from 3 different longitudinal cohorts: Alzheimer's Disease Neuroimaging Initiative, Open Access Series of Imaging Studies-3, and Group of Neuropsychological Studies of the Canary Islands (GENIC). In total, we generated 7548 images to simulate the access of a scan per subject every 6 months in these cohorts. We evaluated the quality of the synthetic images using six quantitative measurements and a qualitative assessment by an experienced neuroradiologist with state-of-the-art results. The assumption of linear brain decay was accurate in these cohorts (R2 ∈ [.924,.940]). The experimental results show that the proposed methodology can produce anatomically plausible aging predictions that can be used to enhance longitudinal datasets. Compared to deep learning-based generative methods, diffeomorphic registration is more likely to preserve the anatomy of the different structures of the brain, which makes it more appropriate for its use in clinical applications. The proposed methodology is able to efficiently simulate anatomically plausible 3D MRI scans of brain aging of healthy subjects from two images scanned at two different time points.

 Simulating prospective magnetic resonance imaging (MRI) scans from a given individual brain image is challenging, as it requires accounting for canonical changes in aging and/or disease progression while also considering the individual brain’s current status and unique characteristics. While current deep generative models can produce high-resolution anatomically accurate templates for population-wide studies, they cannot predict future aging trajectories for individuals. In this study, we introduce Individualized Brain Synthesis (InBrainSyn), a framework that synthesizes high resolution longitudinal MRI scans that simulate subject-specific neurodegeneration in both Alzheimer’s disease (AD) and normal aging. InBrainSyn uses a parallel transport algorithm to adapt the population-level aging trajectories learned by a generative deep template network to synthesize subject-specific aging. Diffeomorphic registration guarantees the anatomical plausibility of the synthesized images by design. Experimentally, we find that our model faithfully simulates both individualized aging and inter-cohort generations capturing neuroanatomical transitions between normal aging and AD on the OASIS-3 dataset. Overall, given only a single scan from an individual, InBrainSyn produces high-fidelity and anatomically plausible longitudinal aging trajectories of synthesized 3D spatiotemporal T1w MRI scans. The code is shared at https://github.com/Fjr9516/InBrainSyn. 

Osteoporosis, characterized by reduced bone mineral density and micro-architectural degeneration, significantly enhances fracture-risk. There are several viable methods for trabecular bone micro-imaging, which widely vary in terms of technology, reconstruction principle, spatial resolution, and acquisition time. We have performed an excised cadaveric bone specimen study to evaluate different computed tomography (CT)-imaging modalities for trabecular bone micro-structural analysis. Excised cadaveric bone specimens from the distal radius were scanned using micro-CT and four in vivo CT imaging modalities: high-resolution peripheral quantitative computed tomography (HR-pQCT), dental cone beam CT (CBCT), whole-body multi-row detector CT (MDCT), and extremity CBCT. A new algorithm was developed to optimize soft thresholding parameters for individual in vivo CT modalities for computing quantitative bone volume fraction maps. Finally, agreement of trabecular bone micro-structural measures, derived from different in vivo CT imaging, with reference measures from micro-CT imaging was examined. Observed values of most trabecular measures, including trabecular bone volume, network area, transverse and plate-rod micro-structure, thickness, and spacing, for in vivo CT modalities were higher than their micro-CT-based reference values. In general, HR-pQCT-based trabecular bone measures were closer to their reference values as compared to other in vivo CT modalities. Despite large differences in observed values of measures among modalities, high linear correlation (r ∈ [0.94 0.99]) was found between micro-CT and in vivo CT-derived measures of trabecular bone volume, transverse and plate micro-structural volume, and network area. All HR-pQCT-derived trabecular measures, except the erosion index, showed high correlation (r ∈ [0.91 0.99]). The plate-width measure showed a higher correlation (r ∈ [0.72 0.91]) among in vivo and micro-CT modalities than its counterpart binary plate-rod characterization-based measure erosion index (r ∈ [0.65 0.81]). Although a strong correlation was observed between micro-structural measures from in vivo and micro-CT imaging, large shifts in their values for in vivo modalities warrant proper scanner calibration prior to adopting in multi-site and longitudinal studies.

Tractography has become an indispensable part of brain connectivity studies. However, it is currently facing problems with reliability. In particular, a substantial amount of nerve fiber reconstructions (streamlines) in tractograms produced by state-of-the-art tractography methods are anatomically implausible. To address this problem, tractogram filtering methods have been developed to remove faulty connections in a postprocessing step. This study takes a closer look at one such method, Spherical-deconvolution Informed Filtering of Tractograms (SIFT), which uses a global optimization approach to improve the agreement between the remaining streamlines after filtering and the underlying diffusion magnetic resonance imaging data. SIFT is not suitable for judging the compliance of individual streamlines with the acquired data since its results depend on the size and composition of the surrounding tractogram. To tackle this problem, we propose applying SIFT to randomly selected tractogram subsets in order to retrieve multiple assessments for each streamline. This approach makes it possible to identify streamlines with very consistent filtering results, which were used as pseudo-ground truths for training classifiers. The trained classifier is able to distinguish the obtained groups of complying and non-complying streamlines with the acquired data with an accuracy above 80%.

One of the main issues of current tractography methods is their high false-positive rate. Tractogram filtering is an option for removing false positive streamlines from tractography data in a post-processing step. In this paper, we train a deep neural network for filtering tractography data in which every streamline of a tractogram is classified as plausible, implausible or inconclusive. For this, we use four different tractogram filtering strategies as supervisors, whose outputs are combined to obtain the classification labels for the streamlines. We assessed the importance of different features of the streamlines for performing this classification task, including the coordinates of the streamlines, diffusion data, landmarks, T1-weighted information and a brain parcellation. We found that the streamline coordinates are the most relevant, followed by the diffusion data, in this particular classification task.

Tractography aims at describing the most likely neural fiber paths in white matter. A general issue of current tractography methods is their large false-positive rate. An approach to deal with this problem is tractogram filtering in which anatomically implausible streamlines are discarded as a post-processing step after tractography. In this chapter, we review the main approaches and methods from the literature that are relevant for the application of tractogram filtering. Moreover, we give a perspective on the central challenges for the development of new methods, including modern machine learning techniques, in this field in the next few years.

Perceptual organisation techniques aim at mimicking the human visual system for extracting salient information from noisy images. Tensor voting has been one of the most versatile of those methods, with many different applications both in computer vision and medical image analysis. Its strategy consists in propagating local information encoded through tensors by means of perception-inspired rules. Although it has been used for more than a decade, there are still many unsolved theoretical issues that have made it challenging to apply it to more problems, especially in analysis of medical images.

The main aim of this chapter is to review the current state of the research in tensor voting, to summarise its present challenges, and to describe the new trends that we foresee will drive the research in this field in the next few years. Also, we discuss extensions of tensor voting that could lead to potential performance improvements and that could make it suitable for further medical applications.

Tensor Voting is a technique that uses perceptual rules to group points in a set of input data. Its main advantage lies in its ability to robustly extract geometrical shapes like curves and surfaces from point clouds even in noisy scenarios. Following the original formulation this is achieved by exploiting the relative positioning of those points with respect to each other. Having this in mind, it is not a straight forward task to apply original tensor voting to greyscale images. Due to the underlying voxel grid, digital images have all data measurements at regularly sampled positions. For that reason, the pure spatial position of data points relative to each other does not provide useful information unless one considers the measured intensity value in addition to that. To account for that, previous approaches of employing tensor voting to scalar images have followed mainly two ideas. One is to define a subset of voxels that are likely to resemble a desired structure like curves or surfaces in the original image in a preprocessing step and to use only those points for initialisation in tensor voting. In other methods, the encoding step is modified e.g. by using estimations of local orientations for initialisation. In contrast to these approaches, another idea is to embed all information given as input, that is position in combination with intensity value, into a 4D space and perform classic tensor voting on that. In doing so, it is neither necessary to rely on a preprocessing step for estimating local orientation features nor is it needed to employ assumptions within the encoding step as all data points are initialised with unit ball tensors. Alternatively, the intensity dimension could be partially included by considering it in the weighting function of tensor voting while still employing 3D tensors for the voting. Considering the advantage of a shorter computation time for the latter approach, it is of interest to investigate the differences between these two approaches. Although different methods have employed an ND implementation of tensor voting before, the actual interpretation of its output, that is the estimation of a local hyper surface at each point, depends on the actual application at hand. As we are especially interested in the analysis of blood vessels in CT angiography data, we study the feasibility of detecting tubular structures and the estimation of their orientation totally within the proposed framework and also compare the two mentioned approaches with a special focus on these aspects. In this chapter we first provide the formulation of both approaches followed by the application-specific interpretations of the shape of 4D output tensors. Based on that, we compare the information inferred by both methods from both synthetic and medical image data focusing on the application of blood vessel analysis.

mong the various diffusion MRI techniques, diffusion ten-sor imaging (DTI) is still most commonly used in clinicalpractice in order to investigate connectivity and fibre anatomyin the human brain. Besides its apparent advantages of a shortacquisition time and noise robustness compared to other tech-niques, it suffers from its major weakness of assuming a sin-gle fibre model in each voxel. This constitutes a problem forDTI fibre tracking algorithms in regions with crossing fibres.Methods approaching this problem in a postprocessing stepemploy diffusion-like techniques to correct the directional in-formation. We propose an extension of tensor voting in whichinformation from voxels with a single fibre is used to inferorientation distributions in multi fibre voxels. The method isable to resolve multiple fibre orientations by clustering tensorvotes instead of adding them up. Moreover, a new vote cast-ing procedure is proposed which is appropriate even for smallneighbourhoods. To account for the locality of DTI data, weuse a small neighbourhood for distributing information at atime, but apply the algorithm iteratively to close larger gaps.The method shows promising results in both synthetic casesand for processing DTI-data of the human brain.

This paper focuses on predicting a single step of streamline tractography from diffusion magnetic resonance imaging data by using different predictors based on neural networks. We train 18 different classifiers in order to assess the effect of including neighbourhood information in the learning step or as a post processing step. Moreover, the performance using four different post processing approaches as well as the variation of the number of classes resulting in a total of 60 experimental configurations are assessed. Further, a comparison to 12 regression-based networks is performed and the effect of including several streamline steps in the network input is investigated. All networks are trained and tested on the ISMRM 2015 tractography challenge data. Our results do not indicate a clear improvement when using neighbouring data (regardless if it used as an input or as a post processing). Also, the linear interpolation of the diffusion data does not outperform the less expensive nearest neighbour approach. As opposed to that, using a linear model on top of the output of the classifiers is beneficial and—in combination with at least 200 classes—resulted in a similar performance as the regression approach. Finally, providing the networks with additional curvature information led to a clear improvement of prediction performance. Our analysis of accuracy based on average angular errors suggests that also considering spatial location in the learning step might further improve machine learning-based streamline tractography algorithms.

Among the various diffusion MRI techniques, diffusion tensor imaging (DTI) is still most commonly used in clinical practice in order to investigate connectivity and fibre anatomy in the human brain. Besides its apparent advantages of a short acquisition time and noise robustness compared to other techniques, it suffers from its major weakness of assuming a single fibre model in each voxel. This constitutes a problem for DTI fibre tracking algorithms in regions with crossing fibres. Methods approaching this problem in a postprocessing step employ diffusion-like techniques to correct the directional information. We propose an extension of tensor voting in which information from voxels with a single fibre is used to infer orientation distributions in multi fibre voxels. The method is able to resolve multiple fibre orientations by clustering tensor votes instead of adding them up. Moreover, a new vote casting procedure is proposed which is appropriate even for small neighbourhoods. To account for the locality of DTI data, we use a small neighbourhood for distributing information at a time, but apply the algorithm iteratively to close larger gaps. The method shows promising results in both synthetic cases and for processing DTI-data of the human brain.

Background As bone microstructure is known to impact bone strength, the aim of this in vitro study was to evaluate if the emerging photon-counting detector computed tomography (PCD-CT) technique may be used for measurements of trabecular bone structures like thickness, separation, nodes, spacing and bone volume fraction. Methods Fourteen cubic sections of human radius were scanned with two multislice CT devices, one PCD-CT and one energy-integrating detector CT (EID-CT), using micro-CT as a reference standard. The protocols for PCD-CT and EID-CT were those recommended for inner- and middle-ear structures, although at higher mAs values: PCD-CT at 450 mAs and EID-CT at 600 (dose equivalent to PCD-CT) and 1000 mAs. Average measurements of the five bone parameters as well as dispersion measurements of thickness, separation and spacing were calculated using a three-dimensional automated region growing (ARG) algorithm. Spearman correlations with micro-CT were computed. Results Correlations with micro-CT, for PCD-CT and EID-CT, ranged from 0.64 to 0.98 for all parameters except for dispersion of thickness, which did not show a significant correlation (p = 0.078 to 0.892). PCD-CT had seven of the eight parameters with correlations rho > 0.7 and three rho > 0.9. The dose-equivalent EID-CT instead had four parameters with correlations rho > 0.7 and only one rho > 0.9. Conclusions In this in vitro study of radius specimens, strong correlations were found between trabecular bone structure parameters computed from PCD-CT data when compared to micro-CT. This suggests that PCD-CT might be useful for analysing bone microstructure in the peripheral human skeleton.

Trabecular bone structure has been shown to impact bone strength and fracture risk. In vitro, this structure can be measured by micro-computed tomography (micro-CT). For clinical use, it would be valuable if multi-slice computed tomography (MSCT) could be used to analyse trabecular bone structure. One important step in the analysis is image volume segmentation. Previous segmentation techniques have either been computer resource intensive or produced sub-optimal results when used on MSCT data. This paper proposes a new segmentation method that tries to balance good results against computational complexity. Material. Fourteen human radius specimens where scanned with MSCT and segmented using the proposed method as well as two segmentation methods previously used to segment trabecular bone (Otsu and Automated Region Growing (ARG)). The proposed method (named FCH) uses a combination of feature space clustering, edge detection and hysteresis thresholding. For evaluation, we computed correlations with the reference method micro-CT for 7 structure parameters and measured segmentation time. Results. Correlations with micro-CT were highest for FCH in 3 cases, highest for ARG in 3 cases, and in general lower for Otsu. Both FCH and ARG had correlations higher than 0.80 for all parameters, except for trabecular thickness and trabecular termini. FCH was 60 times slower than Otsu, but 5 times faster than ARG. Discussion. The high correlations with micro-CT suggest that with a suitable segmentation method it might be possible to analyse trabecular bone structure using MSCT-machines. The proposed segmentation method may represent a useful balance between speed and accuracy.

Stiffness and shear moduli of human trabecular bone may be analyzed in vivo by finite element (FE) analysis from image data obtained by clinical imaging equipment such as high resolution peripheral quantitative computed tomography (HR-pQCT). In clinical practice today, this is done in the peripheral skeleton like the wrist and heel. In this cadaveric bone study, fourteen bone specimens from the wrist were imaged by two dental cone beam computed tomography (CBCT) devices and one HR-pQCT device as well as by dual energy X-ray absorptiometry (DXA). Histomorphometric measurements from micro-CT data were used as gold standard. The image processing was done with an in-house developed code based on the automated region growing (ARG) algorithm. Evaluation of how well stiffness (Young's modulus E3) and minimum shear modulus from the 12, 13, or 23 could be predicted from the CBCT and HR-pQCT imaging data was studied and compared to FE analysis from the micro-CT imaging data. Strong correlations were found between the clinical machines and micro-CT regarding trabecular bone structure parameters, such as bone volume over total volume, trabecular thickness, trabecular number and trabecular nodes (varying from 0.79 to 0.96). The two CBCT devices as well as the HR-pQCT showed the ability to predict stiffness and shear, with adjusted R-2-values between 0.78 and 0.92, based on data derived through our in-house developed code based on the ARG algorithm. These findings indicate that clinically used CBCT may be a feasible method for clinical studies of bone structure and mechanical properties in future osteoporosis research.

Objectives. The aim of this study was to evaluate the possibility of estimating the biomechanical properties of trabecular bone through finite element simulations by using dental cone beam computed tomography data. Study Design. Fourteen human radius specimens were scanned in 3 cone beam computed tomography devices: 3-D Accuitomo 80 (J. Morita MFG., Kyoto, Japan), NewTom 5 G (QR Verona, Verona, Italy), and Verity (Planmed, Helsinki, Finland). The imaging data were segmented by using 2 different methods. Stiffness (Young modulus), shear moduli, and the size and shape of the stiffness tensor were studied. Corresponding evaluations by using micro-CT were regarded as the reference standard. Results. The 3-D Accuitomo 80 (J. Morita MFG., Kyoto, Japan) showed good performance in estimating stiffness and shear moduli but was sensitive to the choice of segmentation method. Newtom 5 G (QR Verona, Verona, Italy) and Verity (Planmed, Helsinki, Finland) yielded good correlations, but they were not as strong as Accuitomo 80 U. Morita MFG., Kyoto, Japan). The cone beam computed tomography devices overestimated both stiffness and shear compared with the micro-CT estimations. Conclusions. Finite element-based calculations of biomechanics from cone beam computed tomography data are feasible, with strong correlations for the Accuitomo 80 scanner a. Morita MFG., Kyoto, Japan) combined with an appropriate segmentation method. Such measurements might be useful for predicting implant survival by in vivo estimations of bone properties.

Bone strength depends on both mineral content and bone structure. Measurements of bone microstructure on specimens can be performed by micro-CT. In vivo measurements are reliably performed by high-resolution peripheral computed tomography (HR-pQCT) using dedicated software. In previous studies from our research group, trabecular bone properties on CT data of defatted specimens from many different CT devices have been analyzed using an Automated Region Growing (ARG) algorithm-based code, showing strong correlations to micro-CT. The aim of the study was to validate the possibility of segmenting and measuring trabecular bone structure from clinical CT data of fresh-frozen human wrist specimens. Data from micro-CT was used as reference. The hypothesis was that the ARG-based in-house built software could be used for such measurements. HR-pQCT image data at two resolutions (61 and 82 µm isotropic voxels) from 23 fresh-frozen human forearms were analyzed. Correlations to micro-CT were strong, varying from 0.72 to 0.99 for all parameters except trabecular termini and nodes. The bone volume fraction had correlations varying from 0.95 to 0.98 but was overestimated compared to micro-CT, especially at the lower resolution. Trabecular separation and spacing were the most stable parameters with correlations at 0.80-0.97 and mean values in the same range as micro-CT. Results from this in vitro study show that an ARG-based software could be used for segmenting and measuring 3D trabecular bone structure from clinical CT data of fresh-frozen human wrist specimens using micro-CT data as reference. Over-and underestimation of several of the bone structure parameters must however be taken into account.

Purpose: The aim of this work was to quantify the extent of lipid-rich necrotic core (LRNC) and intraplaque hemorrhage (IPH) in atherosclerotic plaques. Methods: Patients scheduled for carotid endarterectomy underwent four-point Dixon and T1-weighted magnetic resonance imaging (MRI) at 3 Tesla. Fat and R2* maps were generated from the Dixon sequence at the acquired spatial resolution of 0.60×0.60×0.70mm voxel size. MRI and three-dimensional (3D) histology volumes of plaques were registered. The registration matrix was applied to segmentations denoting LRNC and IPH in 3D histology to split plaque volumes in regions with and without LRNC and IPH. Results: Five patients were included. Regarding volumes of LRNC identified by 3D histology, the average fat fraction by MRI was significantly higher inside LRNC than outside: 12.64±0.2737% versus 9.294±0.1762% (mean±standard error of the mean [SEM]; P<0.001). The same was true for IPH identified by 3D histology, R2* inside versus outside IPH was: 71.81±1.276 s-1 versus 56.94±0.9095 s-1 (mean±SEM; P<0.001). There was a strong correlation between the cumulative fat and the volume of LRNC from 3D histology (R2=0.92) as well as between cumulative R2* and IPH (R2=0.94). Conclusion: Quantitative mapping of fat and R2* from Dixon MRI reliably quantifies the extent of LRNC and IPH.

Measurement of anisotropy of trabecular bone has clinical relevance in osteoporosis. In this study, anisotropy measurements of 15 trabecular bone biopsies from the radius estimated by different fabric tensors on images acquired through cone beam computed tomography (CBCT) and micro computed tomography (micro-CT) were compared. The results show that the generalized mean intercept length (MIL) tensor performs better than the global gray-scale structure tensor, especially when the von Mises-Fisher kernel is applied.  Also, the generalized MIL tensor yields consistent results between the two scanners. These results suggest that this tensor is appropriate for estimating anisotropy in images acquired in vivo through CBCT.

Vesselness filters aim at enhancing tubular structures in medical images. The most popular vesselness filters are based on eigenanalyses of the Hessian matrix computed at different scales. However, Hessian-based methods have well-known limitations, most of them related to the use of second order derivatives. In this paper, we propose an alternative strategy in which ring-like patterns are sought in the local orientation distribution of the gradient. The method takes advantage of symmetry properties of ring-like patterns in the spherical harmonics domain. For bright vessels, gradients not pointing towards the center are filtered out from every local neighborhood in a first step. The opposite criterion is used for dark vessels. Afterwards, structuredness, evenness and uniformness measurements are computed from the power spectrum in spherical harmonics of both the original and the half-zeroed orientation distribution of the gradient. Finally, the features are combined into a single vesselness measurement. Alternatively, a structure tensor that is suitable for vesselness can be estimated before the analysis in spherical harmonics. The two proposed methods are called Ring Pattern Detector (RPD) and Filtered Structure Tensor (FST) respectively. Experimental results with computed tomography angiography data show that the proposed filters perform better compared to the state-of-the-art.

We recently proposed a method for estimating vesselness based on detection of ring patterns in the local distribution ofthe gradient. This method has a better performance than other state-of-the-art algorithms. However, the original implementation of the method makes use of the spherical harmonics transform locally, which is time consuming. In this paper we propose an equivalent formulation of the method based on higher-order tensors. A linear mapping between the spherical harmonics transform and higher-order orientation tensors is used in order to reduce the complexity of the method. With the new implementation, the analysis of computed tomography angiography data can be performed 2.6 times faster compared with the original implementation.

The apparent stiffness tensor is an important mechanical parameter for characterizing trabecular bone. Previous studies have modeled this parameter as a function of mechanical properties of the tissue, bone density and a second-order fabric tensor, which encodes both anisotropy and orientation of trabecular bone. Although these models yield strong correlations between observed and predicted stiffness tensors, there is still space for reducing accuracy errors.In this paper we propose a model that uses fourth-order instead of second-order fabric tensors. First, the totally symmetric part of the stiffness tensor is assumed proportional to the fourth-order fabric tensor in the logarithmic scale. Second, the asymmetric part of the stiffness tensor is derived from relationships among components of the harmonic tensor decomposition of the stiffness tensor. The mean intercept length (MIL), generalized MIL (GMIL) and global structure tensor fourth-order were computed from images acquired through micro computed tomography of 264 specimens of the femur. The predicted tensors were compared to the stiffness tensors computed by using the micro finite element method (micro-FE), which was considered as the gold standard, yielding strong correlations (R^2 above 0.962). The GMIL tensor yielded the best results among the tested fabric tensors. The Frobenius error, geodesic error and the error of the norm were reduced by applying the proposed model by 3.75%, 0.07% and 3.16%, respectively compared to the model by Zysset and Curnier (1995) with the second-order MIL tensor. From the results, fourth-order fabric tensors are a good alternative to the more expensive micro-FE stiffness predictions.

This paper proposes an automatic segmentation method of vessel walls that combines an implicit 3D model of the vessels and a total curvature penalizer in a level set evolution scheme. First, the lumen is segmented by alternating a model-guided level set evolution and a recalculation of the model itself. Second, the level set of the lumen is evolved with a term that aims at penalizing the total curvature and with a prior that forces the outer layer of the vessel towards the outside of the lumen. The model term is deactivated during this step. Finally, in a third step, the model term is reactivated in order to impose a smooth change of the radius along the vessel. Once the two segmentations have been computed, stenoses are detected and quantified at the thickest locations of the segmented vessel wall. Preliminary results show that the proposed method compares favorably with respect to the state-of-the-art both for synthetic and real CTA datasets.

Introduction: In tractography, redundancy poses a significant challenge, often resulting in tractograms that include anatomically implausible streamlines or those that fail to represent the brain's white matter architecture accurately. Current filtering methods aim to refine tractograms by addressing these issues, but they lack a unified measure of redundancy and can be computationally demanding. Methods: We propose a novel framework to quantify tractogram redundancy based on filtering tractogram subsets without endorsing a specific filtering algorithm. Our approach defines redundancy based on the anatomical plausibility and diffusion signal representation of streamlines, establishing both lower and upper bounds for the number of false-positive streamlines and the tractogram redundancy. Results: We applied this framework to tractograms from the Human Connectome Project, using geometrical plausibility and statistical methods informed by the streamlined attributes and ensemble consensus. Our results establish bounds for the tractogram redundancy and the false-discovery rate of the tractograms. Conclusion: This study advances the understanding of tractogram redundancy and supports the refinement of tractography methods. Future research will focus on further validating the proposed framework and exploring tractogram compression possibilities.

IntroductionAcoustic radiation is one of the most important white matter fiber bundles of the human auditory system. However, segmenting the acoustic radiation is challenging due to its small size and proximity to several larger fiber bundles. TractSeg is a method that uses a neural network to segment some of the major fiber bundles in the brain. This study aims to train TractSeg to segment the core of acoustic radiation. MethodsWe propose a methodology to automatically extract the acoustic radiation from human connectome data, which is both of high quality and high resolution. The segmentation masks generated by TractSeg of nearby fiber bundles are used to steer the generation of valid streamlines through tractography. Only streamlines connecting the Heschl's gyrus and the medial geniculate nucleus were considered. These streamlines are then used to create masks of the core of the acoustic radiation that is used to train the neural network of TractSeg. The trained network is used to automatically segment the acoustic radiation from unseen images. ResultsThe trained neural network successfully extracted anatomically plausible masks of the core of the acoustic radiation in human connectome data. We also applied the method to a dataset of 17 patients with unilateral congenital ear canal atresia and 17 age- and gender-paired controls acquired in a clinical setting. The method was able to extract 53/68 acoustic radiation in the dataset acquired with clinical settings. In 14/68 cases, the method generated fragments of the acoustic radiation and completely failed in a single case. The performance of the method on patients and controls was similar. DiscussionIn most cases, it is possible to segment the core of the acoustic radiations even in images acquired with clinical settings in a few seconds using a pre-trained neural network.

Hypothesis Unilateral congenital conductive hearing impairment in ear canal atresia leads to atrophy of the gray matter of the contralateral primary auditory cortex or changes in asymmetry pattern if left untreated in childhood. Background Unilateral ear canal atresia with associated severe conductive hearing loss results in deteriorated sound localization and difficulties in understanding of speech in a noisy environment. Cortical atrophy in the Heschl's gyrus has been reported in acquired sensorineural hearing loss but has not been studied in unilateral conductive hearing loss. Methods We obtained T1w and T2w FLAIR MRI data from 17 subjects with unilateral congenital ear canal atresia and 17 matched controls. Gray matter volume and thickness were measured in the Heschl's gyrus using Freesurfer. Results In unilateral congenital ear canal atresia, Heschl's gyrus exhibited cortical thickness asymmetry (right thicker than left, corrected p = 0.0012, mean difference 0.25 mm), while controls had symmetric findings. Gray matter volume and total thickness did not differ from controls with normal hearing. Conclusion We observed cortical thickness asymmetry in congenital unilateral ear canal atresia but no evidence of contralateral cortex atrophy. Further research is needed to understand the implications of this asymmetry on central auditory processing deficits.

Objectives

The objective is to investigate if unilateral conductive hearing loss in ear canal atresia without hearing treatment in childhood leads to cortical reorganization in functional connectivity and gray matter morphology.

Design

A prospective study including 18 patients with unilateral congenital ear canal atresia age- and gender-matched with normal hearing controls were examined with audiometry, T1, T2 FLAIR and resting state functional magnetic resonance imaging. The thickness and volume of 68 cortical regions were computed from the anatomical images, and seed-based correlation analysis was performed on the resting state functional data for 6 auditory cortical regions of interest.

Results

No statistically significant differences were seen when applying correction for multiple comparisons. However, trends were observed (uncorrected p<0.05) showing larger cortical volumes of the right precuneus cortex in the atretic group (p=0.014), as well as increased functional connectivity of this region coupled to the planum polare in the right hemisphere (p<0.02). The right side precuneus cortex volume was also the most important variable for distinguishing between patients and controls. Cortical volumes of right primary motor cortex (p=0.034) and right somatosensory cortex (p=0.043) were also larger in the atretic group. No differences were observed in the primary auditory cortices’ volume or thickness.

Conclusion

No differences were found within the primary auditory cortices in cortical thickness or volume, which might reflect childhood plasticity with increased bilateral cortical representation of the normal ear, or cross-modal plasticity with stimuli from other senses. Morphology and functional connectivity pattern indicate increased integration of visual and auditory input in unilateral atresia, although future studies are required to support these findings.

ObjectiveSurvival Rate Prediction (SRP) is a valuable tool to assist in the clinical diagnosis and treatment planning of lung cancer patients. In recent years, deep learning (DL) based methods have shown great potential in medical image processing in general and SRP in particular. This study proposes a fully-automated method for SRP from computed tomography (CT) images, which combines an automatic segmentation of the tumor and a DL-based method for extracting rotational-invariant features. MethodsIn the first stage, the tumor is segmented from the CT image of the lungs. Here, we use a deep-learning-based method that entails a variational autoencoder to provide more information to a U-Net segmentation model. Next, the 3D volumetric image of the tumor is projected onto 2D spherical maps. These spherical maps serve as inputs for a spherical convolutional neural network that approximates the log risk for a generalized Cox proportional hazard model. ResultsThe proposed method is compared with 17 baseline methods that combine different feature sets and prediction models using three publicly-available datasets: Lung1 (n=422), Lung3 (n=89), and H&N1 (n=136). We observed comparable C-index scores compared to the best-performing baseline methods in a 5-fold cross-validation on Lung1 (0.59 +/- 0.03 vs. 0.62 +/- 0.04). In comparison, it slightly outperforms all methods in inter-data set evaluation (0.64 vs. 0.63). The best-performing method from the first experiment reduced its performance to 0.61 and 0.62 for Lung3 and H&N1, respectively. DiscussionThe experiments suggest that the performance of spherical features is comparable with previous approaches, but they generalize better when applied to unseen datasets. That might imply that orientation-independent shape features are relevant for SRP. The performance of the proposed method was very similar, using manual and automatic segmentation methods. This makes the proposed model useful in cases where expert annotations are not available or difficult to obtain.

Structural connectomics derives subject-specific brain connectivity from diffusion-weighted MRI and has potential as a biomarker for clinical Parkinson’s disease (PD) detection. In this study, we applied probabilistic tractography (iFOD2) to derive different types of connectomes and analyzed group discriminability between PD patients and healthy controls from the Parkinson's Progression Markers Initiative (PPMI) dataset (n = 233). Particular emphasis was placed on the streamline filtering stage with SIFT2 and the comparison of different connectivity metrics, including streamline count, fractional anisotropy (FA), axial diffusivity (AD), mean diffusivity (MD), and radial diffusivity (RD). We performed a three-level analysis comprising 1) connection-level statistical analysis, 2) graph theory measures at the node and whole-brain levels, and 3) classification using support vector machines (SVM) and graph neural networks. We did not find any statistical difference at any level after correction for multiple comparisons. Also, the classifiers performed poorly with AUC values close to chance levels. However, we found differences between filtered and unfiltered tractograms at the node level, which could suggest that filtering is a necessary step in structural connectivity analyses. Our findings suggest that structural connectivity analysesfor PD are highly sensitive to specific pipeline configurations and fine-tuning.

Brain tractography involves mapping diffusion-weighted images (DWI) onto streamlines representing neural fibre bundles. Recent research avenues have framed tractography into a reinforcement learning (RL) framework with actor-critic models. However, previous RL-based methods may compromise geometrical relations between the input (DWI) and output (tractogram).

More specifically, 3D rotations applied to the input of RL-based tractography are not adequately reflected in the output, indicating a lack of SO3 equivariance. This study aims to restore the equivariance present in previous non-learning-based methods (e.g., iFOD2 from MRtrix3) to RL-based tractography.

To achieve this, we introduce SO3 equivariant and invariant components for the actors (direction prediction model) and critics (q-value prediction model), respectively. We employ an SE3-equivariant transformer as the next direction prediction function. The fact that both the input DWI and the output directional update can be represented as spherical tensors and transform under representations of SO3 makes this formulation a natural fit for the present problem. 

Another benefit of RL-based tractography is that incorporating local neighbourhoods can help mitigate well-known tractography problems (e.g. kissing, crossing, fanning). Our proposed algorithm extracts neighbourhood information as a predefined graph with spherical signals on the nodes.

The contribution of this work is threefold. First, we discuss rotational equivariance in streamlined tractography on a theoretical level. Second, we propose a method that combines RL-based tractography with an equivariant model. Third, we evaluate the equivariance of the proposed method both locally and globally.

The apparent stiffness tensor is relevant for characterizing trabecular bone quality. Previous studies have used morphology-stiffness relationships for estimating the apparent stiffness tensor. In this paper, we propose to train spherical convolutional neural networks (SphCNNs) to estimate this tensor. Information of the edges, trabecular thickness, and spacing are summarized in functions on the unitary sphere used as inputs for the SphCNNs. The concomitant dimensionality reduction makes it possible to train neural networks on relatively small datasets. The predicted tensors were compared to the stiffness tensors computed by using the micro-finite element method (mu FE), which was considered as the gold standard, and models based on fourth-order fabric tensors. Combining edges and trabecular thickness yields significant improvements in the accuracy compared to the methods based on fourth-order fabric tensors. From the results, SphCNNs are promising for replacing the more expensive mu FE stiffness estimations.

Purpose: Suspected fractures are among the most common reasons for patients to visit emergency departments and often can be difficult to detect and analyze them on film scans. Therefore, we aimed to design a Deep Learning-based tool able to help doctors in diagnosis of bone fractures, following the hierarchical classification proposed by the Arbeitsgemeinschaft fur Osteosynthesefragen (AO) Foundation and the Orthopaedic Trauma Association (OTA). Methods: 2453 manually annotated images of proximal femur were used for the classification in different fracture types (1133 Unbroken femur, 570 type A, 750 type B). Secondly, the A type fractures were further classified into the types A1, A2, A3. Two approaches were implemented: the first is a fine-tuned InceptionV3 convolutional neural network (CNN), used as a baseline for our own proposed approach; the second is a multistage architecture composed by successive CNNs in cascade, perfectly suited to the hierarchical structure of the AO/OTA classification. Gradient Class Activation Maps (Grad-CAM) where used to visualize the most relevant areas of the images for classification. The averaged ability of the CNN was measured with accuracy, area under receiver operating characteristics curve (AUC), recall, precision and F1-score. The averaged ability of the orthopedists with and without the help of the CNN was measured with accuracy and Cohen's Kappa coefficient. Results: We obtained an averaged accuracy of 0.86 (CI 0.84-0.88) for three classes classification and 0.81 (CI 0.79-0.82) for five classes classification. The average accuracy improvement of specialists was 14 % with and without the CAD (Computer Assisted Diagnosis) system. Conclusion: We showed the potential of using a CAD system based on CNN for improving diagnosis accuracy and for helping students with a lower level of expertise. We started our work with proximal femur fractures and we aim to extend it to all bone segments further in the future, in order to implement a tool that could be used in every-day hospital routine.

In recent years, bone fracture detection and classification has been a widely discussed topic and many researchers have proposed different methods to tackle this problem. Despite this, a universal approach able to classify all the fractures in the human body has not yet been defined. We aim to analyze and evaluate a selection of papers, chosen according to their representative approach, where the authors applied different deep learning techniques to classify bone fractures, in order to select the strengths of each of them and try to delineate a generalized strategy. Each study is summarized and evaluated using a radar graph with six values: area under the curve (AUC), test accuracy, sensitivity, specificity, dataset size and labelling reliability. Plus, we defined the key points which should be taken into account when trying to accomplish this purpose and we compared each study with our baseline. In recent years, deep learning and, in particular, the convolution neural network (CNN), has achieved results comparable to those of humans in bone fracture classification. Adopting a correct generalization, we are reasonably sure that a computer-aided diagnosis (CAD) system, correctly designed to assist doctors, would save a considerable amount of time and would limit the number of wrong diagnoses.

Segmentation of diagnostic radiography images using deep learning is progressively expanding, which sets demands on the accessibility, availability, and accuracy of the software tools used. This study aimed at evaluating the performance of a segmentation model for digital breast tomosynthesis (DBT), with the use of computer-simulated breast anatomy. We have simulated breast anatomy and soft tissue breast lesions, by utilizing a model approach based on the Perlin noise algorithm. The obtained breast phantoms were projected and reconstructed into DBT slices using a publicly available open-source reconstruction method. Each lesion was then segmented using two approaches: 1. the Segment Anything Model (SAM), a publicly available AI-based method for image segmentation and 2. manually by three human observers. The lesion area in each slice was compared to the ground truth area, derived from the binary mask of the lesion model. We found similar performance between SAM and manual segmentation. Both SAM and the observers performed comparably in the central slice (mean absolute relative error compared to the ground truth and standard deviation SAM: 4 ± 3 %, observers: 3 ± 3 %). Similarly, both SAM and the observers overestimated the lesion area in the peripheral reconstructed slices (mean absolute relative error and standard deviation SAM: 277 ± 190 %, observers: 295 ± 182 %). We showed that 3D voxel phantoms can be used for evaluating different segmentation methods. In preliminary comparison, tumor segmentation in simulated DBT images using SAM open-source method showed a similar performance as manual tumor segmentation.

The 3D breast ultrasound is a radiation-free and effective imaging technology for breast tumor diagnosis. However, checking the 3D breast ultrasound is time-consuming compared to mammograms. To reduce the workload of radiologists, we proposed a 2.5D deep learning-based breast ultrasound tumor classification system. First, we used the pre-trained STU-Net to finetune and segment the tumor in 3D. Then, we fine-tuned the DenseNet-121 for classification using the 10 slices with the biggest tumoral area and their adjacent slices. The Tumor Detection, Segmentation, and Classification on Automated 3D Breast Ultrasound (TDSC-ABUS) MICCAI Challenge 2023 dataset was used to train and validate the performance of the proposed method. Compared to a 3D convolutional neural network model and radiomics, our proposed method has better performance.