Subcortical Segmentation Research Articles

Analysis of the extent of limbic system changes in multiple sclerosis using FreeSurfer and voxel-based morphometry approaches.

Background and purposeThe limbic brain is involved in diverse cognitive, emotional, and autonomic functions. Injury of the various parts of the limbic system have been correlated with clinical deficits in MS. The purpose of this study was to comprehensively examine different regions of the subcortical limbic system to assess the extent of damage within this entire system as it may be pertinent in correlating with specific aspects of cognitive and behavioral dysfunction in MS by using a fully automated, unbiased segmentation approach.MethodsSixty-seven subjects were included in this study, including 52 with multiple sclerosis (MS) and 15 healthy controls. Only patients with stable MS disease, without any relapses, MRI activity, or disability progression were included. Subcortical limbic system segmentation was performed using the FreeSurfer pipeline ScLimbic, which provides volumes for fornix, mammillary bodies, hypothalamus, septal nuclei, nucleus accumbens, and basal forebrain. Hippocampus and anterior thalamic nuclei were added as additional components of the limbic circuitry, also segmented through FreeSurfer. Whole limbic region mask was generated by combining these structures and used for Voxel-based morphometry (VBM) analysis.ResultsThe mean [95% confidence interval] of the total limbic system volume was lower (0.22% [0.21–0.23]) in MS compared to healthy controls (0.27%, [0.25–0.29], p < .001). Pairwise comparisons of individual limbic regions between MS and controls was significant in the nucleus accumbens (0.046%, [0.043–0.050] vs. 0.059%, [0.051–0.066], p = .005), hypothalamus (0.062%, [0.059–0.065] vs. 0.074%, [0.068–0.081], p = .001), basal forebrain (0.038%, [0.036–0.040] vs. 0.047%, [0.042–0.051], p = .001), hippocampus (0.47%, [0.45–0.49] vs. 0.53%, [0.49–0.57], p = .004), and anterior thalamus (0.077%, [0.072–0.082] vs. 0.093%, [0.084–0.10], p = .001) after Bonferroni correction. Volume of several limbic regions was significantly correlated with T2 lesion burden and brain parenchymal fraction (BPF). Multiple regression model showed minimal influence of BPF on limbic brain volume and no influence of other demographic and disease state variables. VBM analysis showed cluster differences in the fornix and anterior thalamic nuclei at threshold p < 0.05 after adjusting for covariates but the results were insignificant after family-wise error corrections.ConclusionsThe results show evidence that brain volume loss is fairly extensive in the limbic brain. Given the significance of the limbic system in many disease states including MS, such volumetric analyses can be expanded to studying cognitive and emotional disturbances in larger clinical trials. FreeSurfer ScLimbic pipeline provided an efficient and reliable methodology for examining many of the subcortical structures related to the limbic brain.

Subcortical and hippocampal brain segmentation in 5-year-old children: Validation of FSL-FIRST and FreeSurfer against manual segmentation.

Developing accurate subcortical volumetric quantification tools is crucial for neurodevelopmental studies, as they could reduce the need for challenging and time‐consuming manual segmentation. In this study, the accuracy of two automated segmentation tools, FSL‐FIRST (with three different boundary correction settings) and FreeSurfer, were compared against manual segmentation of the hippocampus and subcortical nuclei, including the amygdala, thalamus, putamen, globus pallidus, caudate and nucleus accumbens, using volumetric and correlation analyses in 80 5‐year‐olds.Both FSL‐FIRST and FreeSurfer overestimated the volume on all structures except the caudate, and the accuracy varied depending on the structure. Small structures such as the amygdala and nucleus accumbens, which are visually difficult to distinguish, produced significant overestimations and weaker correlations with all automated methods. Larger and more readily distinguishable structures such as the caudate and putamen produced notably lower overestimations and stronger correlations. Overall, the segmentations performed by FSL‐FIRST's default pipeline were the most accurate, whereas FreeSurfer's results were weaker across the structures.In line with prior studies, the accuracy of automated segmentation tools was imperfect with respect to manually defined structures. However, apart from amygdala and nucleus accumbens, FSL‐FIRST's agreement could be considered satisfactory (Pearson correlation > 0.74, intraclass correlation coefficient (ICC) > 0.68 and Dice score coefficient (DSC) > 0.87) with highest values for the striatal structures (putamen, globus pallidus, caudate) (Pearson correlation > 0.77, ICC > 0.87 and DSC > 0.88, respectively). Overall, automated segmentation tools do not always provide satisfactory results, and careful visual inspection of the automated segmentations is strongly advised.

Contextual-wise discriminative feature extraction and robust network learning for subcortical structure segmentation

Contextual-wise discriminative feature extraction and robust network learning for subcortical structure segmentation

Subcortical segmentation of the fetal brain in 3D ultrasound using deep learning

The quantification of subcortical volume development from 3D fetal ultrasound can provide important diagnostic information during pregnancy monitoring. However, manual segmentation of subcortical structures in ultrasound volumes is time-consuming and challenging due to low soft tissue contrast, speckle and shadowing artifacts. For this reason, we developed a convolutional neural network (CNN) for the automated segmentation of the choroid plexus (CP), lateral posterior ventricle horns (LPVH), cavum septum pellucidum et vergae (CSPV), and cerebellum (CB) from 3D ultrasound. As ground-truth labels are scarce and expensive to obtain, we applied few-shot learning, in which only a small number of manual annotations (n = 9) are used to train a CNN. We compared training a CNN with only a few individually annotated volumes versus many weakly labelled volumes obtained from atlas-based segmentations. This showed that segmentation performance close to intra-observer variability can be obtained with only a handful of manual annotations. Finally, the trained models were applied to a large number (n = 278) of ultrasound image volumes of a diverse, healthy population, obtaining novel US-specific growth curves of the respective structures during the second trimester of gestation.

Comparison of cortical and subcortical structural segmentation methods in Alzheimer’s disease: A statistical approach

BackgroundAutomated segmentation methods are developed to help with the segmentation of different brain areas. However, their reliability has yet to be fully investigated. To have a more comprehensive understanding of the distribution of changes in Alzheimer’s disease (AD), as well as investigating the reliability of different segmentation methods, in this study we compared volumes of cortical and subcortical brain segments, using HIPS, volBrain, CAT and BrainSuite automated segmentation methods between AD, mild cognitive impairment (MCI) and healthy controls (HC). MethodsA total of 182 MRI images were taken from the minimal interval resonance imaging in Alzheimer's disease (MIRIAD; 22 AD and 22 HC) and the Alzheimer’s disease neuroimaging initiative database (ADNI; 43 AD, 50 MCI and 45 HC) datasets. Statistical methods were used to compare different groups as well as the correlation between different methods. ResultsThe two methods of volBrain and CAT showed a strong correlation (p’s < 0.035 Bonferroni corrected for multiple comparisons). The two methods, however, showed no significant correlation with BrainSuite (p’s > 0.820 Bonferroni corrected). Furthermore, BrainSuite did not follow the same trend as the other three methods and only HIPS, volBrain and CAT showed strong conformity with the past literature with strong correlation with mini mental state examination (MMSE) scores. ConclusionOur results showed that automated segmentation methods HIPS, volBrain and CAT can be used in the classification of HC, AD and MCI. This is an indication that such methods can be used to inform researchers and clinicians of underlying mechanisms and progression of AD.

3D hemisphere-based convolutional neural network for whole-brain MRI segmentation.

Whole-brain segmentation is a crucial pre-processing step for many neuroimaging analyses pipelines. Accurate and efficient whole-brain segmentations are important for many neuroimage analysis tasks to provide clinically relevant information. Several recently proposed convolutional neural networks (CNN) perform whole brain segmentation using individual 2D slices or 3D patches as inputs due to graphical processing unit (GPU) memory limitations, and use sliding windows to perform whole brain segmentation during inference. However, these approaches lack global and spatial information about the entire brain and lead to compromised efficiency during both training and testing. We introduce a 3D hemisphere-based CNN for automatic whole-brain segmentation of T1-weighted magnetic resonance images of adult brains. First, we trained a localization network to predict bounding boxes for both hemispheres. Then, we trained a segmentation network to segment one hemisphere, and segment the opposing hemisphere by reflecting it across the mid-sagittal plane. Our network shows high performance both in terms of segmentation efficiency and accuracy (0.84 overall Dice similarity and 6.1 mm overall Hausdorff distance) in segmenting 102 brain structures. On multiple independent test datasets, our method demonstrated a competitive performance in the subcortical segmentation task and a high consistency in volumetric measurements of intra-session scans.

Зв’язок SWI-інтенсивності та легкої паркінсонічної симптоматики в пацієнтів із хворобою дрібних судин

Актуальність. Легка паркінсонічна симптоматика (ЛПС) (mild parkinsonian sіgns) може бути проявом нейродегенеративного процесу, результатом дисциркуляторної енцефалопатії або проявлятися як наслідок цих двох процесів, що перебігають паралельно та посилюють один одного. Мета дослідження: виявити, чи існує взаємозв’язок між зміною магнітної сприйнятливості підкіркових структур головного мозку та наявністю легкої паркінсонічної симптоматики в пацієнтів із хронічною ішeмією мозку. Матеріали та методи. Для дослідження були відібрані 82 пацієнти, у яких було діагностовано дисциркуляторну енцефалопатію та виключено нейродегенеративні захворювання. Усім пацієнтам було виконано магнітно­резонансну томографію (МРТ) у послідовностях T1­WI, T2­WI, TIRM, DWI, SWI. Наявність MPS діагностувалася за допомогою третьої секції шкали UPDRS. Результати. За допомогою U­критерію Манна — Уїтні було виявлено, що пацієнти з ЛПС мали вірогідно вищі рівні магнітної сприйнятливості підкіркових структур і, як наслідок, більшу кількість гіпоінтенсивних вокселів на МРТ­зображеннях у SWI­послідовності. Програмне забезпечення 3D Slicer використовувалося для сегментації підкіркових структур за рівнями інтенсивності сигналу. У пацієнтів із ЛПС було виявлено більшу кількість гіпоінтенсивних вокселів у блідих кулях, шкаралупі та чорних субстанціях обох півкуль. Після врегулювання за віком та балом шкали Fazekas було виявлено, що ЛПС найбільше асоційована із збільшенням кількості вокселів вираженої гіпоінтенсивності (SI &lt; 75) у правій блідій кулі (OR 1,29, CI 1,08–1,55). Висновки. Підвищення рівнів гіпоінтенсивності підкіркових структур, що виявляються на SWI МРТ­знімках, можуть бути асоційовані з ЛПС у пацієнтів із хворобою дрібних судин головного мозку.

The role of multimodal MRI in mild cognitive impairment and Alzheimer's disease.

Mild cognitive impairment (MCI) is a prodromal stage of Alzheimer's disease (AD), where neurodegeneration is not as considerable, thereby potentially increasing the effect of treatments. Therefore, highly sensitive and specific classification of subjects with MCI is necessary, where various MRI modalities have displayed promise. Structural, diffusion, and resting-state (RS) functional MRI analyses were performed on the AD (n = 26), MCI (n = 5), and healthy control (HC) (n = 14) group. Structural analysis was performed via voxel-based morphometry (VBM) and volumetric subcortical segmentation analysis. Fractional anisotropy and mean diffusivity were estimated during the diffusion analysis. RS analysis investigated seed-based functional connectivity. Classification via support vector machine was performed to evaluate which MRI modality most accurately differentiated the groups. Multiple linear regression was conducted to evaluate the MRI modalities correlation with clinical assessment scores. Classification of MCI and HC displayed highest accuracy based on diffusion MRI, which besides demonstrated high correlation with clinical scores. Classification was equally accurate in AD, when using VBM or diffusion tensor imaging measures. Yet, more variance was explained by VBM measures in the clinical assessment scores of the AD group. This study highlights the potential of diffusion MRI in differentiating MCI from HC and AD. However, the results need to be interpreted with caution as sample size and artifacts in the MRI data probably influenced the results.

DeepQSMSeg: A Deep Learning-based Sub-cortical Nucleus Segmentation Tool for Quantitative Susceptibility Mapping.

Deep brain nuclei are closely related to the pathogenesis of neurodegenerative diseases. Automatic segmentation for brain nuclei plays a significant role in aging and disease-related assessment. Quantitative susceptibility mapping (QSM), as a novel MRI imaging technique, attracts increasing attention in deep gray matter (DGM) nuclei-related research and diagnosis. This paper proposes DeepQSMSeg, a deep learning-based end-to-end tool, to segment five pairs of DGM structures from QSM images. The proposed model is based on a 3D encoder-decoder fully convolutional neural network. For concentrating network on the target regions, spatial and channel attention modules are adopted in both encoder and decoder stages. Dice loss is combined with focal loss to alleviate the imbalance of ROI classes. The result shows that our method can segment DGM structures from QSM images precisely, rapidly and reliably. Comparing with ground truth, the average Dice coefficient for all ROIs in the test dataset achieved 0.872±0.053, and Hausdorff distance was 2.644±2.917 mm. Finally, an age-related susceptibility development model was used to confirm the reliability of DeepQSMSeg in aging and disease-related studies.Clinical Relevance-Accurate and automatic segmentation tool for sub-cortical regions in QSM can significantly alleviate the pressure of radiologists. It can also accelerate the progress of related research and clinical translation.

Automated subcortical volume estimation from 2D MRI in epilepsy and implications for clinical trials

PurposeMost techniques used for automatic segmentation of subcortical brain regions are developed for three-dimensional (3D) MR images. MRIs obtained in non-specialist hospitals may be non-isotropic and two-dimensional (2D). Automatic segmentation of 2D images may be challenging and represents a lost opportunity to perform quantitative image analysis. We determine the performance of a modified subcortical segmentation technique applied to 2D images in patients with idiopathic generalised epilepsy (IGE).MethodsVolume estimates were derived from 2D (0.4 × 0.4 × 3 mm) and 3D (1 × 1x1mm) T1-weighted acquisitions in 31 patients with IGE and 39 healthy controls. 2D image segmentation was performed using a modified FSL FIRST (FMRIB Integrated Registration and Segmentation Tool) pipeline requiring additional image reorientation, cropping, interpolation and brain extraction prior to conventional FIRST segmentation. Consistency between segmentations was assessed using Dice coefficients and volumes across both approaches were compared between patients and controls. The influence of slice thickness on consistency was further assessed using 2D images with slice thickness increased to 6 mm.ResultsAll average Dice coefficients showed excellent agreement between 2 and 3D images across subcortical structures (0.86–0.96). Most 2D volumes were consistently slightly lower compared to 3D volumes. 2D images with increased slice thickness showed lower agreement with 3D images with lower Dice coefficients (0.55–0.83). Significant volume reduction of the left and right thalamus and putamen was observed in patients relative to controls across 2D and 3D images.ConclusionAutomated subcortical volume estimation of 2D images with a resolution of 0.4 × 0.4x3mm using a modified FIRST pipeline is consistent with volumes derived from 3D images, although this consistency decreases with an increased slice thickness. Thalamic and putamen atrophy has previously been reported in patients with IGE. Automated subcortical volume estimation from 2D images is feasible and most reliable at using in-plane acquisitions greater than 1 mm x 1 mm and provides an opportunity to perform quantitative image analysis studies in clinical trials.

Cortical and Subcortical Brain Volumes Partially Mediate the Association between Dietary Composition and Behavioral Disinhibition: A UK Biobank Study.

Behavioral disinhibition is observed to be an important characteristic of many neurodevelopmental and psychiatric disorders. Recent studies have linked dietary quality to levels of behavioral inhibition. However, it is currently unclear whether brain factors might mediate this. The current study investigates whether cortical and subcortical brain volumes mediate part of the association between dietary composition and behavioral disinhibition. A total of 15,258 subjects from the UK Biobank project were included in the current study. Dietary composition and behavioral disinhibition were based on Principle Component Analyses of self-reported dietary composition). As a further data reduction step, cortical and subcortical volume segmentations were input into an Independent Component Analysis. The resulting four components were used as mediator variables in the main mediation analyses, where behavioral disinhibition served as the outcome variable and dietary components as predictors. Our results show: (1) significant associations between all dietary components and brain volume components; (2) brain volumes are associated with behavioral disinhibition; (3) the mediation models show that part of the variance in behavioral disinhibition explained by dietary components (for healthy diet, restricted diet, and high-fat dairy diet) is mediated through the frontal-temporal/parietal brain volume component. These results are in part confirming our hypotheses and offer a first insight into the underlying mechanisms linking dietary composition, frontal-parietal brain volume, and behavioral disinhibition in the general adult population.

A deep learning toolbox for automatic segmentation of subcortical limbic structures from MRI images

A tool was developed to automatically segment several subcortical limbic structures (nucleus accumbens, basal forebrain, septal nuclei, hypothalamus without mammillary bodies, the mammillary bodies, and fornix) using only a T1-weighted MRI as input. This tool fills an unmet need as there are few, if any, publicly available tools to segment these clinically relevant structures. A U-Net with spatial, intensity, contrast, and noise augmentation was trained using 39 manually labeled MRI data sets. In general, the Dice scores, true positive rates, false discovery rates, and manual-automatic volume correlation were very good relative to comparable tools for other structures. A diverse data set of 698 subjects were segmented using the tool; evaluation of the resulting labelings showed that the tool failed in less than 1% of cases. Test-retest reliability of the tool was excellent. The automatically segmented volume of all structures except mammillary bodies showed effectiveness at detecting either clinical AD effects, age effects, or both. This tool will be publicly released with FreeSurfer (surfer.nmr.mgh.harvard.edu/fswiki/ScLimbic). Together with the other cortical and subcortical limbic segmentations, this tool will allow FreeSurfer to provide a comprehensive view of the limbic system in an automated way.

Language Disorder in Progressive Supranuclear Palsy and Corticobasal Syndrome: Neural Correlates and Detection by the MLSE Screening Tool.

Background: Progressive supranuclear palsy (PSP) and corticobasal syndrome (CBS) affect speech and language as well as motor functions. Clinical and neuropathological data indicate a close relationship between these two disorders and the non-fluent variant of primary progressive aphasia (nfvPPA). We use the recently developed Mini Linguistic State Examination tool (MLSE) to study speech and language disorders in patients with PSP, CBS, and nfvPPA, in combination with structural magnetic resonance imaging (MRI).Methods: Fifty-one patients (PSP N = 13, CBS N = 19, nfvPPA N = 19) and 30 age-matched controls completed the MLSE, the short form of the Boston Diagnostic Aphasia Examination (BDAE), and the Addenbrooke’s Cognitive Examination III. Thirty-eight patients and all controls underwent structural MRI at 3 Tesla, with T1 and T2-weighted images processed by surface-based and subcortical segmentation within FreeSurfer 6.0.0 to extract cortical thickness and subcortical volumes. Morphometric differences were compared between groups and correlated with the severity of speech and language impairment.Results: CBS and PSP patients showed impaired MLSE performance, compared to controls, with a similar language profile to nfvPPA, albeit less severe. All patient groups showed reduced cortical thickness in bilateral frontal regions and striatal volume. PSP and nfvPPA patients also showed reduced superior temporal cortical thickness, with additional thalamic and amygdalo-hippocampal volume reductions in nfvPPA. Multivariate analysis of brain-wide cortical thickness and subcortical volumes with MLSE domain scores revealed associations between performance on multiple speech and language domains with atrophy of left-lateralised fronto-temporal cortex, amygdala, hippocampus, putamen, and caudate.Conclusions: The effect of PSP and CBS on speech and language overlaps with nfvPPA. These three disorders cause a common anatomical pattern of atrophy in the left frontotemporal language network and striatum. The MLSE is a short clinical screening tool that can identify the language disorder of PSP and CBS, facilitating clinical management and patient access to future clinical trials.

MSGSE-Net: Multi-scale guided squeeze-and-excitation network for subcortical brain structure segmentation

Convolutional neural networks (CNNs) have been achieving remarkable results in medical image segmentation. However, for accurate segmentation of subcortical brain structure in MR images, it is still a challenge due to the ambiguous boundaries, the complex structures, and the various shapes, which limits their clinical application. In this paper, we focus on utilizing multi-scale image contexts and attention mechanisms to improve networks’ ability to learn discriminative feature representation for accurate segmentation and present a novel FCNN architecture called multis-scale guided squeeze-and-excitation network (MSGSE-Net). In particular, we first propose the multi-scale guided squeeze-and-excitation (MSGSE) attention module which can progressively and selectively aggregate discriminative features. In contrast to existing attention modules, the MSGSE module performs an adaptive recalibration that features at different locations of the feature map are recalibrated under the guidance of multi-scale contexts. Then multi-scale spatial attention supervision is adopted to enhance the intra-class homogeneity and inter-class distinction of the attention weights. Moreover, we propose a novel entropy-weighted Dice loss (EDL) to force the network to focus on the ambiguous voxels around the boundaries of subcortical structures. We evaluate the proposed method on two challenging benchmark datasets (the IBSR dataset and the MALC dataset). The experimental results show that our model consistently yields better segmentation performance than several state-of-the-art methods and improves the segmentation Dice score by 1.6% at most compared with baseline method U-Net. Our code is available at https://github.com/neulxlx/MSGSE-Net.

Feasibility of accelerated 3D T1-weighted MRI using compressed sensing: application to quantitative volume measurements of human brain structures.

Scan time reduction is necessary for volumetric acquisitions to improve workflow productivity and to reduce motion artifacts during MRI procedures. We explored the possibility that Compressed Sensing-4 (CS-4) can be employed with 3D-turbo-field-echo T1-weighted (3D-TFE-T1W) sequence without compromising subcortical measurements on clinical 1.5T MRI. Thirty-three healthy volunteers (24 females, 9 males) underwent imaging scans on a 1.5T MRI equipped with a 12-channel head coil. 3D-TFE-T1W for whole-brain coverage was performed with different acceleration factors, including SENSE-2, SENSE-4, CS-4. Freesurfer, FSL's FIRST, and volBrain packages were utilized for subcortical segmentation. All processed data were assessed using the Wilcoxon signed-rank test. The results obtained from SENSE-2 were considered as references. For SENSE-4, the maximum signal-to-noise ratio (SNR) drop was detected in the Accumbens (51.96%). For CS-4, the maximum SNR drop was detected in the Amygdala (10.55%). Since the SNR drop in CS-4 is relatively small, the SNR in all of the subcortical volumes obtained from SENSE-2 and CS-4 are not statistically different (P > 0.05), and their Pearson's correlation coefficients are larger than 0.90. The maximum biases of SENSE-4 and CS-4 were found in the Thalamus with the mean of differences of 1.60ml and 0.18ml, respectively. CS-4 provided sufficient quality of 3D-TFE-T1W images for 1.5T MRI equipped with a 12-channel receiver coil. Subcortical volumes obtained from the CS-4 images are consistent among different post-processing packages.

HybraPD atlas: Towards precise subcortical nuclei segmentation using multimodality medical images in patients with Parkinson disease.

Human brain atlases are essential for research and surgical treatment of Parkinson's disease (PD). For example, deep brain stimulation for PD often requires human brain atlases for brain structure identification. However, few atlases can provide disease‐specific subcortical structures for PD, and most of them are based on T1w and T2w images. In this work, we construct a HybraPD atlas using fused quantitative susceptibility mapping (QSM) and T1w images from 87 patients with PD. The constructed HybraPD atlas provides a series of templates, that is, T1w, GRE magnitude, QSM, R2*, and brain tissue probabilistic maps. Then, we manually delineate a parcellation map with 12 bilateral subcortical nuclei, which are highly related to PD pathology, such as sub‐regions in globus pallidus and substantia nigra. Furthermore, we build a whole‐brain parcellation map by combining existing cortical parcellation and white‐matter segmentation with the proposed subcortical nuclei map. Considering the multimodality of the HybraPD atlas, the segmentation accuracy of each nucleus is evaluated using T1w and QSM templates, respectively. The results show that the HybraPD atlas provides more accurate segmentation than existing atlases. Moreover, we analyze the metabolic difference in subcortical nuclei between PD patients and healthy control subjects by applying the HybraPD atlas to calculate uptake values of contrast agents on positron emission tomography (PET) images. The atlas‐based analysis generates accurate disease‐related brain nuclei segmentation on PET images. The newly developed HybraPD atlas could serve as an efficient template to study brain pathological alterations in subcortical regions for PD research.

Joint super-resolution and synthesis of 1 mm isotropic MP-RAGE volumes from clinical MRI exams with scans of different orientation, resolution and contrast

Most existing algorithms for automatic 3D morphometry of human brain MRI scans are designed for data with near-isotropic voxels at approximately 1 mm resolution, and frequently have contrast constraints as well-typically requiring T1-weighted images (e.g., MP-RAGE scans). This limitation prevents the analysis of millions of MRI scans acquired with large inter-slice spacing in clinical settings every year. In turn, the inability to quantitatively analyze these scans hinders the adoption of quantitative neuro imaging in healthcare, and also precludes research studies that could attain huge sample sizes and hence greatly improve our understanding of the human brain. Recent advances in convolutional neural networks (CNNs) are producing outstanding results in super-resolution and contrast synthesis of MRI. However, these approaches are very sensitive to the specific combination of contrast, resolution and orientation of the input images, and thus do not generalize to diverse clinical acquisition protocols – even within sites. In this article, we present SynthSR, a method to train a CNN that receives one or more scans with spaced slices, acquired with different contrast, resolution and orientation, and produces an isotropic scan of canonical contrast (typically a 1 mm MP-RAGE). The presented method does not require any preprocessing, beyond rigid coregistration of the input scans. Crucially, SynthSR trains on synthetic input images generated from 3D segmentations, and can thus be used to train CNNs for any combination of contrasts, resolutions and orientations without high-resolution real images of the input contrasts. We test the images generated with SynthSR in an array of common downstream analyses, and show that they can be reliably used for subcortical segmentation and volumetry, image registration (e.g., for tensor-based morphometry), and, if some image quality requirements are met, even cortical thickness morphometry. The source code is publicly available at https://github.com/BBillot/SynthSR.

Mapping Structural Connectivity Using Diffusion MRI: Challenges and Opportunities

The effects of the misalignment between image-intensity-based tissue segmentation and brain parcellation on the assignment of streamlines to network nodes: Upper panel: (a) With the application of anatomical constraints (e.g., Ref. 35), streamline endpoints (colored in purple) occur at the GM-WM interface or within the subcortical GM. (b) Due to factors such as discretization of structural labels, many of these endpoints (purple points) do not locate inside the, for example, FreeSurfer parcellation image (blue ribbon)48 and thus are not assigned to a label. (c) The T1 image shown in (a,b) is replaced by the GM partial volume maps derived from intensity-based tissue segmentation.80 (d) A zoom region of (c) illustrates the discrepancy (pointed by arrows) between tissue segmentation and brain parcellation, revealing that streamlines cannot be assigned purely based on the voxels where the endpoints reside. Bottom panel: The discrepancies also present in subcortical GM, and in fact could be even crucial when parcellation images are prepared by transforming an atlas to an individual's space (i.e., the left branch of Fig. 5). This is because the degree of misalignment between subcortical GM segmentations and brain parcellations could be increased by the registration error. As an example: (e) The AAL atlas is coregistered to individuals' data via linear and nonlinear transformations. The background shows a T1-weighted image slice of the subject. (f) The background image is replaced by the GM partial volume map obtained from tissue segmentation. (g) The red arrows point to the considerable misalignment between two images at the thalami regions by Chun-Hung 1666–1682.

Reliability and sensitivity of two whole-brain segmentation approaches included in FreeSurfer – ASEG and SAMSEG

Accurate and reliable whole-brain segmentation is critical to longitudinal neuroimaging studies. We undertake a comparative analysis of two subcortical segmentation methods, Automatic Segmentation (ASEG) and Sequence Adaptive Multimodal Segmentation (SAMSEG), recently provided in the open-source neuroimaging package FreeSurfer 7.1, with regard to reliability, bias, sensitivity to detect longitudinal change, and diagnostic sensitivity to Alzheimer’s disease. First, we assess intra- and inter-scanner reliability for eight bilateral subcortical structures: amygdala, caudate, hippocampus, lateral ventricles, nucleus accumbens, pallidum, putamen and thalamus. For intra-scanner analysis we use a large sample of participants (n = 1629) distributed across the lifespan (age range = 4–93 years) and acquired on a 1.5T Siemens Avanto (n = 774) and a 3T Siemens Skyra (n = 855) scanners. For inter-scanner analysis we use a sample of 24 participants scanned on the day with three models of Siemens scanners: 1.5T Avanto, 3T Skyra and 3T Prisma. Second, we test how each method detects volumetric age change using longitudinal follow up scans (n = 491 for Avanto and n = 245 for Skyra; interscan interval = 1–10 years). Finally, we test sensitivity to clinically relevant change. We compare annual rate of hippocampal atrophy in cognitively normal older adults (n = 20), patients with mild cognitive impairment (n = 20) and Alzheimer’s disease (n = 20). We find that both ASEG and SAMSEG are reliable and lead to the detection of within-person longitudinal change, although with notable differences between age-trajectories for most structures, including hippocampus and amygdala. In summary, SAMSEG yields significantly lower differences between repeated measures for intra- and inter-scanner analysis without compromising sensitivity to changes and demonstrating ability to detect clinically relevant longitudinal changes.

Transductive Transfer Learning for Domain Adaptation in Brain Magnetic Resonance Image Segmentation.

Segmentation of brain images from Magnetic Resonance Images (MRI) is an indispensable step in clinical practice. Morphological changes of sub-cortical brain structures and quantification of brain lesions are considered biomarkers of neurological and neurodegenerative disorders and used for diagnosis, treatment planning, and monitoring disease progression. In recent years, deep learning methods showed an outstanding performance in medical image segmentation. However, these methods suffer from generalisability problem due to inter-centre and inter-scanner variabilities of the MRI images. The main objective of the study is to develop an automated deep learning segmentation approach that is accurate and robust to the variabilities in scanner and acquisition protocols. In this paper, we propose a transductive transfer learning approach for domain adaptation to reduce the domain-shift effect in brain MRI segmentation. The transductive scenario assumes that there are sets of images from two different domains: (1) source—images with manually annotated labels; and (2) target—images without expert annotations. Then, the network is jointly optimised integrating both source and target images into the transductive training process to segment the regions of interest and to minimise the domain-shift effect. We proposed to use a histogram loss in the feature level to carry out the latter optimisation problem. In order to demonstrate the benefit of the proposed approach, the method has been tested in two different brain MRI image segmentation problems using multi-centre and multi-scanner databases for: (1) sub-cortical brain structure segmentation; and (2) white matter hyperintensities segmentation. The experiments showed that the segmentation performance of a pre-trained model could be significantly improved by up to 10%. For the first segmentation problem it was possible to achieve a maximum improvement from 0.680 to 0.799 in average Dice Similarity Coefficient (DSC) metric and for the second problem the average DSC improved from 0.504 to 0.602. Moreover, the improvements after domain adaptation were on par or showed better performance compared to the commonly used traditional unsupervised segmentation methods (FIRST and LST), also achieving faster execution time. Taking this into account, this work presents one more step toward the practical implementation of deep learning algorithms into the clinical routine.