Multi-Atlas Segmentation with Joint Label Fusion

Non-linear registration improves statistical power to detect hippocampal atrophy in aging and dementia

OBJECTIVE

To compare the performance of different methods for determining hippocampal atrophy rates using longitudinal MRI scans in aging and Alzheimer's disease (AD).

BACKGROUND

Quantifying hippocampal atrophy caused by neurodegenerative diseases is important to follow the course of the disease. In dementia, the efficacy of new therapies can be partially assessed by measuring their effect on hippocampal atrophy. In radiotherapy, the quantification of radiation-induced hippocampal volume loss is of interest to quantify radiation damage. We evaluated plausibility, reproducibility and sensitivity of eight commonly used methods to determine hippocampal atrophy rates using test-retest scans.

MATERIALS AND METHODS

Manual, FSL-FIRST, FreeSurfer, multi-atlas segmentation (MALF) and non-linear registration methods (Elastix, NiftyReg, ANTs and MIRTK) were used to determine hippocampal atrophy rates on longitudinal T1-weighted MRI from the ADNI database. Appropriate parameters for the non-linear registration methods were determined using a small training dataset (N = 16) in which two-year hippocampal atrophy was measured using test-retest scans of 8 subjects with low and 8 subjects with high atrophy rates. On a larger dataset of 20 controls, 40 mild cognitive impairment (MCI) and 20  AD patients, one-year hippocampal atrophy rates were measured. A repeated measures ANOVA analysis was performed to determine differences between controls, MCI and AD patients. For each method we calculated effect sizes and the required sample sizes to detect one-year volume change between controls and MCI (NCTRL_MCI) and between controls and AD (NCTRL_AD). Finally, reproducibility of hippocampal atrophy rates was assessed using within-session rescans and expressed as an average distance measure DAve, which expresses the difference in atrophy rate, averaged over all subjects. The same DAve was used to determine the agreement between different methods.

RESULTS

Except for MALF, all methods detected a significant group difference between CTRL and AD, but none could find a significant difference between the CTRL and MCI. FreeSurfer and MIRTK required the lowest sample sizes (FreeSurfer: NCTRL_MCI = 115, NCTRL_AD = 17 with DAve = 3.26%; MIRTK: NCTRL_MCI = 97, NCTRL_AD = 11 with DAve = 3.76%), while ANTs was most reproducible (NCTRL_MCI = 162, NCTRL_AD = 37 with DAve = 1.06%), followed by Elastix (NCTRL_MCI = 226, NCTRL_AD = 15 with DAve = 1.78%) and NiftyReg (NCTRL_MCI = 193, NCTRL_AD = 14 with DAve = 2.11%). Manually measured hippocampal atrophy rates required largest sample sizes to detect volume change and were poorly reproduced (NCTRL_MCI = 452, NCTRL_AD = 87 with DAve = 12.39%). Atrophy rates of non-linear registration methods also agreed best with each other.

DISCUSSION AND CONCLUSION

Non-linear registration methods were most consistent in determining hippocampal atrophy and because of their better reproducibility, methods, such as ANTs, Elastix and NiftyReg, are preferred for determining hippocampal atrophy rates on longitudinal MRI. Since performances of non-linear registration methods are well comparable, the preferred method would mostly depend on computational efficiency.

Age-specific optimization of T1-weighted brain MRI throughout infancy

The infant brain undergoes drastic morphological and functional development during the first year of life. Three-dimensional T1-weighted Magnetic Resonance Imaging (3D T1w-MRI) is a major tool to characterize the brain anatomy, which however, manifests inherently low and rapidly changing contrast between white matter (WM) and gray matter (GM) in the infant brains (0-12 month-old). Despite the prior efforts made to maximize tissue contrast in the neonatal brains (≤1 months), optimization of imaging methods in the rest of the infancy (1-12 months) is not fully addressed, while brains in the latter period exhibit even more challenging contrast. Here, we performed a systematic investigation to improve the contrast between cortical GM and subcortical WM throughout the infancy. We first performed simultaneous T1 and proton density mapping in a normally developing infant cohort at 3T (n = 57). Based on the evolution of T1 relaxation times, we defined three age groups and simulated the relative tissue contrast between WM and GM in each group. Age-specific imaging strategies were proposed according to the Bloch simulation: inversion time (TI) around 800 ms for the 0-3 month-old group, dual TI at 500 ms and 700 ms for the 3-7 month-old group, and TI around 700 ms for 7-12 month-old group, using a centrically encoded 3D-MPRAGE sequence at 3T. Experimental results with varying TIs in each group confirmed improved contrast at the proposed optimal TIs, even in 3-7 month-old infants who had nearly isointense contrast. We further demonstrated the advantage of improved relative contrast in segmenting the neonatal brains using a multi-atlas segmentation method. The proposed age-specific optimization strategies can be easily adapted to routine clinical examinations, and the improved image contrast would facilitate quantitative analysis of the infant brain development.

Brain ventricle parcellation using a deep neural network: Application to patients with ventriculomegaly

Numerous brain disorders are associated with ventriculomegaly, including both neuro-degenerative diseases and cerebrospinal fluid disorders. Detailed evaluation of the ventricular system is important for these conditions to help understand the pathogenesis of ventricular enlargement and elucidate novel patterns of ventriculomegaly that can be associated with different diseases. One such disease is normal pressure hydrocephalus (NPH), a chronic form of hydrocephalus in older adults that causes dementia. Automatic parcellation of the ventricular system into its sub-compartments in patients with ventriculomegaly is quite challenging due to the large variation of the ventricle shape and size. Conventional brain labeling methods are time-consuming and often fail to identify the boundaries of the enlarged ventricles. We propose a modified 3D U-Net method to perform accurate ventricular parcellation, even with grossly enlarged ventricles, from magnetic resonance images (MRIs). We validated our method on a data set of healthy controls as well as a cohort of 95 patients with NPH with mild to severe ventriculomegaly and compared with several state-of-the-art segmentation methods. On the healthy data set, the proposed network achieved mean Dice similarity coefficient (DSC) of 0.895 ± 0.03 for the ventricular system. On the NPH data set, we achieved mean DSC of 0.973 ± 0.02, which is significantly (p < 0.005) higher than four state-of-the-art segmentation methods we compared with. Furthermore, the typical processing time on CPU-base implementation of the proposed method is 2 min, which is much lower than the several hours required by the other methods. Results indicate that our method provides: 1) highly robust parcellation of the ventricular system that is comparable in accuracy to state-of-the-art methods on healthy controls; 2) greater robustness and significantly more accurate results on cases of ventricular enlargement; and 3) a tool that enables computation of novel imaging biomarkers for dilated ventricular spaces that characterize the ventricular system.

Splenomegaly Segmentation on Multi-Modal MRI Using Deep Convolutional Networks

The findings of splenomegaly, abnormal enlargement of the spleen, is a non-invasive clinical biomarker for liver and spleen diseases. Automated segmentation methods are essential to efficiently quantify splenomegaly from clinically acquired abdominal magnetic resonance imaging (MRI) scans. However, the task is challenging due to: 1) large anatomical and spatial variations of splenomegaly; 2) large inter- and intra-scan intensity variations on multi-modal MRI; and 3) limited numbers of labeled splenomegaly scans. In this paper, we propose the Splenomegaly Segmentation Network (SS-Net) to introduce the deep convolutional neural network (DCNN) approaches in multi-modal MRI splenomegaly segmentation. Large convolutional kernel layers were used to address the spatial and anatomical variations, while the conditional generative adversarial networks were employed to leverage the segmentation performance of SS-Net in an end-to-end manner. A clinically acquired cohort containing both T1-weighted (T1w) and T2-weighted (T2w) MRI splenomegaly scans was used to train and evaluate the performance of multi-atlas segmentation (MAS), 2D DCNN networks, and a 3-D DCNN network. From the experimental results, the DCNN methods achieved superior performance to the state-of-the-art MAS method. The proposed SS-Net method has achieved the highest median and mean Dice scores among the investigated baseline DCNN methods.

A Multi-Atlas Label Fusion Tool for Neonatal Brain MRI Parcellation and Quantification

Structure-by-structure analysis, in which the brain magnetic resonance imaging (MRI) is parcellated based on its anatomical units, is widely used to investigate chronological changes in morphology or signal intensity during normal development, as well as to identify the alterations seen in various diseases or conditions. The multi-atlas label fusion (MALF) method is considered a highly accurate parcellation approach, and anticipated for clinical application to quantitatively evaluate early developmental processes. However, the current MALF methods, which are designed for neonatal brain segmentations, are not widely available. In this study, we developed a T1-weighted, neonatal, multi-atlas repository and integrated it into the MALF-based brain segmentation tools in the cloud-based platform, MRICloud. The cloud platform ensures users instant access to the advanced MALF tool for neonatal brains, with no software or installation requirements for the client. The Web platform by braingps.mricloud.org will eliminate the dependence on a particular operating system (eg, Windows, Macintosh, or Linux) and the requirement for high computational performance of the user's computers. The MALF-based, fully automated, image parcellation could achieve excellent agreement with manual parcellation, and the whole and regional brain volumes quantified through this method demonstrated developmental trajectories comparable to those from a previous publication. This solution will make the latest MALF tools readily available to users, with minimum barriers, and will expedite and accelerate advancements in developmental neuroscience research, neonatology, and pediatric neuroradiology.

Dilated Dense U-Net for Infant Hippocampus Subfield Segmentation

Accurate and automatic segmentation of infant hippocampal subfields from magnetic resonance (MR) images is an important step for studying memory related infant neurological diseases. However, existing hippocampal subfield segmentation methods were generally designed based on adult subjects, and would compromise performance when applied to infant subjects due to insufficient tissue contrast and fast changing structural patterns of early hippocampal development. In this paper, we propose a new fully convolutional network (FCN) for infant hippocampal subfield segmentation by embedding the dilated dense network in the U-net, namely DUnet. The embedded dilated dense network can generate multi-scale features while keeping high spatial resolution, which is useful in fusing the low-level features in the contracting path with the high-level features in the expanding path. To further improve the performance, we group every pair of convolutional layers with one residual connection in the DUnet, and obtain the Residual DUnet (ResDUnet). Experimental results show that our proposed DUnet and ResDUnet improve the average Dice coefficient by 2.1 and 2.5% for infant hippocampal subfield segmentation, respectively, when compared with the classic 3D U-net. The results also demonstrate that our methods outperform other state-of-the-art methods.

Volumetric comparison of hippocampal subfields extracted from 4-minute accelerated vs. 8-minute high-resolution T2-weighted 3T MRI scans

The hippocampus has been widely studied using neuroimaging, as it plays an important role in memory and learning. However, hippocampal subfield information is difficult to capture by standard magnetic resonance imaging (MRI) techniques. To facilitate morphometric study of hippocampal subfields, ADNI introduced a high resolution (0.4 mm in plane) T2-weighted turbo spin-echo sequence that requires 8 min. With acceleration, the protocol can be acquired in 4 min. We performed a comparative study of hippocampal subfield volumes using standard and accelerated protocols on a Siemens Prisma 3T MRI in an independent sample of older adults that included 10 cognitively normal controls, 9 individuals with subjective cognitive decline, 10 with mild cognitive impairment, and 6 with a clinical diagnosis of Alzheimer's disease (AD). The Automatic Segmentation of Hippocampal Subfields (ASHS) software was used to segment 9 primary labeled regions including hippocampal subfields and neighboring cortical regions. Intraclass correlation coefficients were computed for reliability tests between 4 and 8 min scans within and across the four groups. Pairwise group analyses were performed, covaried for age, sex and total intracranial volume, to determine whether the patterns of group differences were similar using 4 vs. 8 min scans. The 4 and 8 min protocols, analyzed by ASHS segmentation, yielded similar volumetric estimates for hippocampal subfields as well as comparable patterns of differences between study groups. The accelerated protocol can provide reliable imaging data for investigation of hippocampal subfields in AD-related MRI studies and the decreased scan time may result in less vulnerability to motion.

The effect of automated landmark identification on morphometric analyses

Morphometric analysis of anatomical landmarks allows researchers to identify specific morphological differences between natural populations or experimental groups, but manually identifying landmarks is time-consuming. We compare manually and automatically generated adult mouse skull landmarks and subsequent morphometric analyses to elucidate how switching from manual to automated landmarking will impact morphometric analysis results for large mouse (Mus musculus) samples (n = 1205) that represent a wide range of 'normal' phenotypic variation (62 genotypes). Other studies have suggested that the use of automated landmarking methods is feasible, but this study is the first to compare the utility of current automated approaches to manual landmarking for a large dataset that allows the quantification of intra- and inter-strain variation. With this unique sample, we investigated how switching to a non-linear image registration-based automated landmarking method impacts estimated differences in genotype mean shape and shape variance-covariance structure. In addition, we tested whether an initial registration of specimen images to genotype-specific averages improves automatic landmark identification accuracy. Our results indicated that automated landmark placement was significantly different than manual landmark placement but that estimated skull shape covariation was correlated across methods. The addition of a preliminary genotype-specific registration step as part of a two-level procedure did not substantially improve on the accuracy of one-level automatic landmark placement. The landmarks with the lowest automatic landmark accuracy are found in locations with poor image registration alignment. The most serious outliers within morphometric analysis of automated landmarks displayed instances of stochastic image registration error that are likely representative of errors common when applying image registration methods to micro-computed tomography datasets that were initially collected with manual landmarking in mind. Additional efforts during specimen preparation and image acquisition can help reduce the number of registration errors and improve registration results. A reduction in skull shape variance estimates were noted for automated landmarking methods compared with manual landmarking. This partially reflects an underestimation of more extreme genotype shapes and loss of biological signal, but largely represents the fact that automated methods do not suffer from intra-observer landmarking error. For appropriate samples and research questions, our image registration-based automated landmarking method can eliminate the time required for manual landmarking and have a similar power to identify shape differences between inbred mouse genotypes.

Deep convolutional neural network for segmentation of thoracic organs-at-risk using cropped 3D images

PURPOSE

Automatic segmentation of organs-at-risk (OARs) is a key step in radiation treatment planning to reduce human efforts and bias. Deep convolutional neural networks (DCNN) have shown great success in many medical image segmentation applications but there are still challenges in dealing with large 3D images for optimal results. The purpose of this study is to develop a novel DCNN method for thoracic OARs segmentation using cropped 3D images.

METHODS

To segment the five organs (left and right lungs, heart, esophagus and spinal cord) from the thoracic CT scans, preprocessing to unify the voxel spacing and intensity was first performed, a 3D U-Net was then trained on resampled thoracic images to localize each organ, then the original images were cropped to only contain one organ and served as the input to each individual organ segmentation network. The segmentation maps were then merged to get the final results. The network structures were optimized for each step, as well as the training and testing strategies. A novel testing augmentation with multiple iterations of image cropping was used. The networks were trained on 36 thoracic CT scans with expert annotations provided by the organizers of the 2017 AAPM Thoracic Auto-segmentation Challenge and tested on the challenge testing dataset as well as a private dataset.

RESULTS

The proposed method earned second place in the live phase of the challenge and first place in the subsequent ongoing phase using a newly developed testing augmentation approach. It showed superior-than-human performance on average in terms of Dice scores (spinal cord: 0.893 ± 0.044, right lung: 0.972 ± 0.021, left lung: 0.979 ± 0.008, heart: 0.925 ± 0.015, esophagus: 0.726 ± 0.094), mean surface distance (spinal cord: 0.662 ± 0.248 mm, right lung: 0.933 ± 0.574 mm, left lung: 0.586 ± 0.285 mm, heart: 2.297 ± 0.492 mm, esophagus: 2.341 ± 2.380 mm) and 95% Hausdorff distance (spinal cord: 1.893 ± 0.627 mm, right lung: 3.958 ± 2.845 mm, left lung: 2.103 ± 0.938 mm, heart: 6.570 ± 1.501 mm, esophagus: 8.714 ± 10.588 mm). It also achieved good performance in the private dataset and reduced the editing time to 7.5 min per patient following automatic segmentation.

CONCLUSIONS

The proposed DCNN method demonstrated good performance in automatic OAR segmentation from thoracic CT scans. It has the potential for eventual clinical adoption of deep learning in radiation treatment planning due to improved accuracy and reduced cost for OAR segmentation.

Thalamus Optimized Multi Atlas Segmentation (THOMAS): fast, fully automated segmentation of thalamic nuclei from structural MRI

The thalamus and its nuclei are largely indistinguishable on standard T1 or T2 weighted MRI. While diffusion tensor imaging based methods have been proposed to segment the thalamic nuclei based on the angular orientation of the principal diffusion tensor, these are based on echo planar imaging which is inherently limited in spatial resolution and suffers from distortion. We present a multi-atlas segmentation technique based on white-matter-nulled MP-RAGE imaging that segments the thalamus into 12 nuclei with computation times on the order of 10 min on a desktop PC; we call this method THOMAS (THalamus Optimized Multi Atlas Segmentation). THOMAS was rigorously evaluated on 7T MRI data acquired from healthy volunteers and patients with multiple sclerosis by comparing against manual segmentations delineated by a neuroradiologist, guided by the Morel atlas. Segmentation accuracy was very high, with uniformly high Dice indices: at least 0.85 for large nuclei like the pulvinar and mediodorsal nuclei and at least 0.7 even for small structures such as the habenular, centromedian, and lateral and medial geniculate nuclei. Volume similarity indices ranged from 0.82 for the smaller nuclei to 0.97 for the larger nuclei. Volumetry revealed that the volumes of the right anteroventral, right ventral posterior lateral, and both right and left pulvinar nuclei were significantly lower in MS patients compared to controls, after adjusting for age, sex and intracranial volume. Lastly, we evaluated the potential of this method for targeting the Vim nucleus for deep brain surgery and focused ultrasound thalamotomy by overlaying the Vim nucleus segmented from pre-operative data on post-operative data. The locations of the ablated region and active DBS contact corresponded well with the segmented Vim nucleus. Our fast, direct structural MRI based segmentation method opens the door for MRI guided intra-operative procedures like thalamotomy and asleep DBS electrode placement as well as for accurate quantification of thalamic nuclear volumes to follow progression of neurological disorders.

Registration-based image enhancement improves multi-atlas segmentation of the thalamic nuclei and hippocampal subfields

Magnetic resonance imaging (MRI) is an important tool for analysis of deep brain grey matter structures. However, analysis of these structures is limited due to low intensity contrast typically found in whole brain imaging protocols. Herein, we propose a big data registration-enhancement (BDRE) technique to augment the contrast of deep brain structures using an efficient large-scale non-rigid registration strategy. Direct validation is problematic given a lack of ground truth data. Rather, we validate the usefulness and impact of BDRE for multi-atlas (MA) segmentation on two sets of structures of clinical interest: the thalamic nuclei and hippocampal subfields. The experimental design compares algorithms using T1-weighted 3 T MRI for both structures (and additional 7 T MRI for the thalamic nuclei) with an algorithm using BDRE. As baseline comparisons, a recent denoising (DN) technique and a super-resolution (SR) method are used to preprocess the original 3 T MRI. The performance of each MA segmentation is evaluated by the Dice similarity coefficient (DSC). BDRE significantly improves mean segmentation accuracy over all methods tested for both thalamic nuclei (3 T imaging: 9.1%; 7 T imaging: 15.6%; DN: 6.9%; SR: 16.2%) and hippocampal subfields (3 T T1 only: 8.7%; DN: 8.4%; SR: 8.6%). We also present DSC performance for each thalamic nucleus and hippocampal subfield and show that BDRE can help MA segmentation for individual thalamic nuclei and hippocampal subfields. This work will enable large-scale analysis of clinically relevant deep brain structures from commonly acquired T1 images.

MIST: A multi-resolution parcellation of functional brain networks

The functional architecture of the brain is organized across multiple levels of spatial resolutions, from distributed networks to the localized areas they are made of. A brain parcellation that defines functional nodes at multiple resolutions is required to investigate the functional connectome across these scales. Here we present the Multiresolution Intrinsic Segmentation Template (MIST), a multi-resolution group level parcellation of the cortical, subcortical and cerebellar gray matter. The individual MIST parcellations match other published group parcellations in internal homogeneity and reproducibility and perform very well in real-world application benchmarks. In addition, the MIST parcellations are fully annotated and provide a hierarchical decomposition of functional brain networks across nine resolutions (7 to 444 functional parcels). We hope that the MIST parcellation will accelerate research in brain connectivity across resolutions. Because visualizing multiresolution parcellations is challenging, we provide an interactive web interface to explore the MIST. The MIST is also available through the popular nilearn toolbox.

Convolutional Neural Networks with Template-Based Data Augmentation for Functional Lung Image Quantification

RATIONALE AND OBJECTIVES

We propose an automated segmentation pipeline based on deep learning for proton lung MRI segmentation and ventilation-based quantification which improves on our previously reported methodologies in terms of computational efficiency while demonstrating accuracy and robustness. The large data requirement for the proposed framework is made possible by a novel template-based data augmentation strategy. Supporting this work is the open-source ANTsRNet-a growing repository of well-known deep learning architectures first introduced here.

MATERIALS AND METHODS

Deep convolutional neural network (CNN) models were constructed and trained using a custom multilabel Dice metric loss function and a novel template-based data augmentation strategy. Training (including template generation and data augmentation) employed 205 proton MR images and 73 functional lung MRI. Evaluation was performed using data sets of size 63 and 40 images, respectively.

RESULTS

Accuracy for CNN-based proton lung MRI segmentation (in terms of Dice overlap) was left lung: 0.93 ± 0.03, right lung: 0.94 ± 0.02, and whole lung: 0.94 ± 0.02. Although slightly less accurate than our previously reported joint label fusion approach (left lung: 0.95 ± 0.02, right lung: 0.96 ± 0.01, and whole lung: 0.96 ± 0.01), processing time is <1 second per subject for the proposed approach versus ∼30 minutes per subject using joint label fusion. Accuracy for quantifying ventilation defects was determined based on a consensus labeling where average accuracy (Dice multilabel overlap of ventilation defect regions plus normal region) was 0.94 for the CNN method; 0.92 for our previously reported method; and 0.90, 0.92, and 0.94 for expert readers.

CONCLUSION

The proposed framework yields accurate automated quantification in near real time. CNNs drastically reduce processing time after offline model construction and demonstrate significant future potential for facilitating quantitative analysis of functional lung MRI.

A randomized, double-blind, placebo-controlled trial of lamotrigine for prescription corticosteroid effects on the human hippocampus

In animals, stress and corticosteroid excess are associated with decreases in memory performance and hippocampal volume that may be prevented with agents that decrease glutamate release. Humans also demonstrate changes in memory and hippocampus with corticosteroids. In this report the effects of glutamate-release inhibitor lamotrigine on hippocampal structure and memory were examined in people receiving medically needed prescription corticosteroid therapy. A total of 54 outpatient adults (n = 28 women) receiving chronic (≥ 6 months) oral corticosteroid therapy were randomized to lamotrigine or placebo for 48 weeks. Declarative memory was assessed using the Rey Auditory Verbal Learning Test (RAVLT); structural magnetic resonance imaging (MRI) as well as single-voxel proton MR spectroscopy (¹HMRS) focused on hippocampus were obtained at baseline and week 48. Utilizing a mixed-model approach, structural and biochemical data were examined by separate ANOVAs, and memory was assessed with a multi-level longitudinal model. RAVLT total scores demonstrated significantly better declarative memory performance with lamotrigine than placebo (p = 0.047). Hippocampal subfield volumes were not significantly different between the treatment groups. In summary, lamotrigine was associated with less decline in declarative memory performance than placebo in corticosteroid-treated patients. Findings suggest that, in humans as well as in animal models, glutamate release inhibitors may attenuate some of the effects on the human memory associated with corticosteroids.

Sex differences in estimated brain metabolism in relation to body growth through adolescence

The human brain consumes a disproportionate amount of the body's overall metabolic resources, and evidence suggests that brain and body may compete for substrate during development. Using perfusion MRI from a large cross-sectional cohort, we examined developmental changes of MRI-derived estimates of brain metabolism, in relation to weight change. Nonlinear models demonstrated that, in childhood, changes in body weight were inversely related to developmental age-related changes in brain metabolism. This inverse relationship persisted through early adolescence, after which body and brain metabolism began to decline. Females achieved maximum body growth approximately two years earlier than males, with a correspondingly earlier stabilization of brain metabolism to adult levels. These findings confirm prior findings with positron emission tomography performed in a much smaller cohort, demonstrate that relative brain metabolism can be inferred from noninvasive MRI data, and extend observations on the associations between body growth and brain metabolism to sex differences through adolescence.

Benchmark on Automatic 6-month-old Infant Brain Segmentation Algorithms: The iSeg-2017 Challenge

Accurate segmentation of infant brain magnetic resonance (MR) images into white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) is an indispensable foundation for early studying of brain growth patterns and morphological changes in neurodevelopmental disorders. Nevertheless, in the isointense phase (approximately 6-9 months of age), due to inherent myelination and maturation process, WM and GM exhibit similar levels of intensity in both T1-weighted (T1w) and T2-weighted (T2w) MR images, making tissue segmentation very challenging. Despite many efforts were devoted to brain segmentation, only few studies have focused on the segmentation of 6-month infant brain images. With the idea of boosting methodological development in the community, iSeg-2017 challenge (http://iseg2017.web.unc.edu) provides a set of 6-month infant subjects with manual labels for training and testing the participating methods. Among the 21 automatic segmentation methods participating in iSeg-2017, we review the 8 top-ranked teams, in terms of Dice ratio, modified Hausdorff distance and average surface distance, and introduce their pipelines, implementations, as well as source codes. We further discuss limitations and possible future directions. We hope the dataset in iSeg-2017 and this review article could provide insights into methodological development for the community.

Quantitative vibratory sensation measurement is related to sensory cortical thickness in MS

Objective

Vibratory sensation is a quantifiable measure of physical dysfunction and is often related to spinal cord pathology; however, its association with relevant brain areas has not been fully explored. Our objective was to establish a cortical structural substrate for vibration sensation.

Methods

Eighty‐four individuals with multiple sclerosis (MS) (n = 54 relapsing, n = 30 progressive) and 28 controls participated in vibratory sensation threshold quantification at the great toe and a 3T MRI evaluating volume of the thalamus and cortical thickness primary and secondary sensory cortices.

Results

After controlling for age, sex, and disability level, vibratory sensation thresholds were significantly related to cortical thickness of the anterior cingulate (P = 0.041), parietal operculum (P = 0.022), and inferior frontal gyrus pars operculum (P = 0.044), pars orbitalis (P = 0.007), and pars triangularis (P = 0.029). Within the progressive disease subtype, there were significant relationships between vibratory sensation and thalamic volume (P = 0.039) as well as reduced inferior frontal gyrus pars operculum (P = 0.014) and pars orbitalis (P = 0.005) cortical thickness.

Conclusions

The data show significant independent relationships between quantitative vibratory sensation and measures of primary and secondary sensory cortices. Quantitative clinical measurement of vibratory sensation reflects pathological changes in spatially distinct brain areas and may supplement information captured by brain atrophy measures. Without overt relapses, monitoring decline in progressive forms of MS has proved challenging; quantitative clinical assessment may provide a tool to examine pathological decline in this cohort. These data suggest that quantitative clinical assessment may be a reliable way to examine pathological decline and have broader relevance to progressive forms of MS.

Brain structure segmentation in the presence of multiple sclerosis lesions

Intensity-based multi-atlas segmentation strategies have shown to be particularly successful in segmenting brain images of healthy subjects. However, in the same way as most of the methods in the state of the art, their performance tends to be affected by the presence of MRI visible lesions, such as those found in multiple sclerosis (MS) patients. Here, we present an approach to minimize the effect of the abnormal lesion intensities on multi-atlas segmentation. We propose a new voxel/patch correspondence model for intensity-based multi-atlas label fusion strategies that leads to more accurate similarity measures, having a key role in the final brain segmentation. We present the theory of this model and integrate it into two well-known fusion strategies: Non-local Spatial STAPLE (NLSS) and Joint Label Fusion (JLF). The experiments performed show that our proposal improves the segmentation performance of the lesion areas. The results indicate a mean Dice Similarity Coefficient (DSC) improvement of 1.96% for NLSS (3.29% inside and 0.79% around the lesion masks) and, an improvement of 2.06% for JLF (2.31% inside and 1.42% around lesions). Furthermore, we show that, with the proposed strategy, the well-established preprocessing step of lesion filling can be disregarded, obtaining similar or even more accurate segmentation results.

•

We present an approach to improve multi-atlas brain parcellation of MS patients.

•

We integrate our model into 2 well-known segmentation strategies.

•

Our model improves the segmentation on the lesion areas.

•

The improvement on the lesion areas is also reflected in the global performance.

•

With our model, lesion filling can be omitted, obtaining at least similar results.

Effect of disease-modifying therapies on subcortical gray matter atrophy in multiple sclerosis

BACKGROUND

The effects of disease-modifying therapies (DMTs) on region-specific brain atrophy in multiple sclerosis (MS) are unclear.

OBJECTIVE

To determine the effects of higher versus lower efficacy DMTs on rates of brain substructure atrophy in MS.

METHODS

A non-randomized, observational cohort of people with MS followed with annual brain magnetic resonance imaging (MRI) was evaluated retrospectively. Whole brain, subcortical gray matter (GM), cortical GM, and cerebral white matter (WM) volume fractions were obtained. DMTs were categorized as higher (DMT-H: natalizumab and rituximab) or lower (DMT-L: interferon-beta and glatiramer acetate) efficacy. Follow-up epochs were analyzed if participants had been on a DMT for ⩾6 months prior to baseline and had at least one follow-up MRI while on DMTs in the same category.

RESULTS

A total of 86 DMT epochs (DMT-H: n = 32; DMT-L: n = 54) from 78 participants fulfilled the study inclusion criteria. Mean follow-up was 2.4 years. Annualized rates of thalamic (-0.15% vs -0.81%; p = 0.001) and putaminal (-0.27% vs -0.73%; p = 0.001) atrophy were slower during DMT-H compared to DMT-L epochs. These results remained significant in multivariate analyses including demographics, clinical characteristics, and T2 lesion volume.

CONCLUSION

DMT-H treatment may be associated with slower rates of subcortical GM atrophy, especially of the thalamus and putamen. Thalamic and putaminal volumes are promising imaging biomarkers in MS.

GAS: A genetic atlas selection strategy in multi-atlas segmentation framework

Multi-Atlas based Segmentation (MAS) algorithms have been successfully applied to many medical image segmentation tasks, but their success relies on a large number of atlases and good image registration performance. Choosing well-registered atlases for label fusion is vital for an accurate segmentation. This choice becomes even more crucial when the segmentation involves organs characterized by a high anatomical and pathological variability. In this paper, we propose a new genetic atlas selection strategy (GAS) that automatically chooses the best subset of atlases to be used for segmenting the target image, on the basis of both image similarity and segmentation overlap. More precisely, the key idea of GAS is that if two images are similar, the performances of an atlas for segmenting each image are similar. Since the ground truth of each atlas is known, GAS first selects a predefined number of similar images to the target, then, for each one of them, finds a near-optimal subset of atlases by means of a genetic algorithm. All these near-optimal subsets are then combined and used to segment the target image. GAS was tested on single-label and multi-label segmentation problems. In the first case, we considered the segmentation of both the whole prostate and of the left ventricle of the heart from magnetic resonance images. Regarding multi-label problems, the zonal segmentation of the prostate into peripheral and transition zone was considered. The results showed that the performance of MAS algorithms statistically improved when GAS is used.

Automated Generation of Radiologic Descriptions on Brain Volume Changes From T1-Weighted MR Images: Initial Assessment of Feasibility

Purpose: To examine the feasibility and potential difficulties of automatically generating radiologic reports (RRs) to articulate the clinically important features of brain magnetic resonance (MR) images.

Materials and Methods: We focused on examining brain atrophy by using magnetization-prepared rapid gradient-echo (MPRAGE) images. The technology was based on multi-atlas whole-brain segmentation that identified 283 structures, from which larger superstructures were created to represent the anatomic units most frequently used in RRs. Through two layers of data-reduction filters, based on anatomic and clinical knowledge, raw images (~10 MB) were converted to a few kilobytes of human-readable sentences. The tool was applied to images from 92 patients with memory problems, and the results were compared to RRs independently produced by three experienced radiologists. The mechanisms of disagreement were investigated to understand where machine–human interface succeeded or failed.

Results: The automatically generated sentences had low sensitivity (mean: 24.5%) and precision (mean: 24.9%) values; these were significantly lower than the inter-rater sensitivity (mean: 32.7%) and precision (mean: 32.2%) of the radiologists. The causes of disagreement were divided into six error categories: mismatch of anatomic definitions (7.2 ± 9.3%), data-reduction errors (11.4 ± 3.9%), translator errors (3.1 ± 3.1%), difference in the spatial extent of used anatomic terms (8.3 ± 6.7%), segmentation quality (9.8 ± 2.0%), and threshold for sentence-triggering (60.2 ± 16.3%).

Conclusion: These error mechanisms raise interesting questions about the potential of automated report generation and the quality of image reading by humans. The most significant discrepancy between the human and automatically generated RRs was caused by the sentence-triggering threshold (the degree of abnormality), which was fixed to z-score >2.0 for the automated generation, while the thresholds by radiologists varied among different anatomical structures.

Hippocampal subfields segmentation in brain MR images using generative adversarial networks

Background

Segmenting the hippocampal subfields accurately from brain magnetic resonance (MR) images is a challenging task in medical image analysis. Due to the small structural size and the morphological complexity of the hippocampal subfields, the traditional segmentation methods are hard to obtain the ideal segmentation result.

Methods

In this paper, we proposed a hippocampal subfields segmentation method using generative adversarial networks. The proposed method can achieve the pixel-level classification of brain MR images by building an UG-net model and an adversarial model and training the two models against each other alternately. UG-net extracts local information and retains the interrelationship features between pixels. Moreover, the adversarial training implements spatial consistency among the generated class labels and smoothens the edges of class labels on segmented region.

Results

The evaluation has performed on the dataset obtained from center for imaging of neurodegenerative diseases (CIND) for CA1, CA2, DG, CA3, Head, Tail, SUB, ERC and PHG in hippocampal subfields, resulting in the dice similarity coefficient (DSC) of 0.919, 0.648, 0.903, 0.673, 0.929, 0.913, 0.906, 0.884 and 0.889 respectively. For the large subfields, such as Head and CA1 of hippocampus, the DSC was increased by 3.9% and 9.03% than state-of-the-art approaches, while for the smaller subfields, such as ERC and PHG, the segmentation accuracy was significantly increased 20.93% and 16.30% respectively.

Conclusion

The results show the improvement in performance of the proposed method, compared with other methods, which include approaches based on multi-atlas, hierarchical multi-atlas, dictionary learning and sparse representation and CNN. In implementation, the proposed method provides better results in hippocampal subfields segmentation.

Multi-atlas cardiac PET segmentation

PURPOSE

We propose a multi-atlas based segmentation method for cardiac PET and SPECT images to deal with the high variability of tracer uptake characteristics in myocardium. In addition, we verify its performance by comparing it to the manual segmentation and single-atlas based approach, using dynamic myocardial PET.

METHODS

Twelve left coronary artery ligated SD rats underwent ([¹⁸F]fluoropentyl) triphenylphosphonium salt PET/CT scans. Atlas-based segmentation is based on the spatial normalized template with pre-defined region-of-interest (ROI) for each anatomical or functional structure. To generate multiple left ventricular (LV) atlases, each LV image was segmented manually and divided into angular segments. The segmentation methods performances were compared in regional count information using leave-one-out cross-validation. Additionally, the polar-maps of kinetic parameters were estimated.

RESULTS

In all images, the highest r² template yielded the lowest root-mean-square error (RMSE) between the source image and the best-matching templates ranged between 0.91-0.97 and 0.06-0.11, respectively. The single-atlas and multi-atlas based ROIs yielded remarkably different perfusion distributions: only the multi-atlas based segmentation showed equivalent high correlation results (r² = 0.92) with the manual segmentation compared with the single-atlas based (r² = 0.88). The high perfusion value underestimation was remarkable in single-atlas based segmentation.

CONCLUSIONS

The main advantage of the proposed multi-atlas based cardiac segmentation method is that it does not require any prior information on the tracer distribution to be incorporated into the image segmentation algorithms. Therefore, the same procedure suggested here is applicable to any other cardiac PET or SPECT imaging agents without modification.

Robust Multicontrast MRI Spleen Segmentation for Splenomegaly Using Multi-Atlas Segmentation

OBJECTIVE

Magnetic resonance imaging (MRI) is an essential imaging modality in noninvasive splenomegaly diagnosis. However, it is challenging to achieve spleen volume measurement from three-dimensional MRI given the diverse structural variations of human abdomens as well as the wide variety of clinical MRI acquisition schemes. Multi-atlas segmentation (MAS) approaches have been widely used and validated to handle heterogeneous anatomical scenarios. In this paper, we propose to use MAS for clinical MRI spleen segmentation for splenomegaly.

METHODS

First, an automated segmentation method using the selective and iterative method for performance level estimation (SIMPLE) atlas selection is used to address the concerns of inhomogeneity for clinical splenomegaly MRI. Then, to further control outliers, semiautomated craniocaudal spleen length-based SIMPLE atlas selection (L-SIMPLE) is proposed to integrate a spatial prior in a Bayesian fashion and guide iterative atlas selection. Last, a graph cuts refinement is employed to achieve the final segmentation from the probability maps from MAS.

RESULTS

A clinical cohort of 55 MRI volumes (28 T1 weighted and 27 T2 weighted) was used to evaluate both automated and semiautomated methods.

CONCLUSION

The results demonstrated that both methods achieved median Dice , and outliers were alleviated by the L-SIMPLE (≍1 min manual efforts per scan), which achieved 0.97 Pearson correlation of volume measurements with the manual segmentation.

SIGNIFICANCE

In this paper, spleen segmentation on MRI splenomegaly using MAS has been performed.

The Relationship Between Cumulative Exogenous Corticosteroid Exposure and Volumes of Hippocampal Subfields and Surrounding Structures

PURPOSE/BACKGROUND

Glucocorticoids are a class of hormones that include naturally occurring cortisol and corticosterone, as well as prescription drugs commonly used to manage inflammatory, autoimmune, and allergic conditions. Adverse effects, including neuropsychiatric symptoms, are common. The hippocampus appears to be especially sensitive to the effects of glucocorticoids. However, to our knowledge, no studies to date have examined hippocampal subfields in humans receiving glucocorticoids. We examined patients on chronic glucocorticoid regimens to determine relationships between dose and duration of treatment, and hippocampal subfields, and related regions volumes.

METHODS/PROCEDURES

The study included adult men and women receiving at least 5 mg daily of prednisone equivalents for at least 6 months. Volumes of brain regions were measured via magnetic resonance imaging. A multivariate general linear model was used for analysis, with brain volumes as dependent variables and age, sex, and cumulative corticosteroid exposure, as predictors.

FINDINGS/RESULTS

The study population consisted of 81 adult outpatients (43 male) on corticosteroids (mean dose, 7.88 mg; mean duration, 76.75 months). Cumulative glucocorticoid exposure was negatively associated with left and right hippocampal dentate gyrus/CA3 volume. In subsequent subgroup analysis, this association held true for the age group older than the median age of 46 years but not for the younger age group.

IMPLICATIONS/CONCLUSIONS

This finding is consistent with previous studies showing detrimental effects of elevated glucocorticoids on the hippocampus but further suggests that the dentate gyrus and CA3 regions are particularly vulnerable to those effects, which is consistent with animal models of chronic stress but has not been previously demonstrated in humans.

Transfer Learning for Image Segmentation by Combining Image Weighting and Kernel Learning

Many medical image segmentation methods are based on the supervised classification of voxels. Such methods generally perform well when provided with a training set that is representative of the test images to the segment. However, problems may arise when training and test data follow different distributions, for example, due to differences in scanners, scanning protocols, or patient groups. Under such conditions, weighting training images according to distribution similarity have been shown to greatly improve performance. However, this assumes that a part of the training data is representative of the test data; it does not make unrepresentative data more similar. We, therefore, investigate kernel learning as a way to reduce differences between training and test data and explore the added value of kernel learning for image weighting. We also propose a new image weighting method that minimizes maximum mean discrepancy (MMD) between training and test data, which enables the joint optimization of image weights and kernel. Experiments on brain tissue, white matter lesion, and hippocampus segmentation show that both kernel learning and image weighting, when used separately, greatly improve performance on heterogeneous data. Here, MMD weighting obtains similar performance to previously proposed image weighting methods. Combining image weighting and kernel learning, optimized either individually or jointly, can give a small additional improvement in performance.

A New Multi-Atlas Registration Framework for Multimodal Pathological Images Using Conventional Monomodal Normal Atlases

Using multi-atlas registration (MAR), information carried by atlases can be transferred onto a new input image for the tasks of region of interest (ROI) segmentation, anatomical landmark detection, and so on. Conventional atlases used in MAR methods are monomodal and contain only normal anatomical structures. Therefore, the majority of MAR methods cannot handle input multimodal pathological images, which are often collected in routine image-based diagnosis. This is because registering monomodal atlases with normal appearances to multimodal pathological images involves two major problems: (1) missing imaging modalities in the monomodal atlases, and (2) influence from pathological regions. In this paper, we propose a new MAR framework to tackle these problems. In this framework, a deep learning based image synthesizers are applied for synthesizing multimodal normal atlases from conventional monomodal normal atlases. To reduce the influence from pathological regions, we further propose a multimodal lowrank approach to recover multimodal normal-looking images from multimodal pathological images. Finally, the multimodal normal atlases can be registered to the recovered multimodal images in a multi-channel way. We evaluate our MAR framework via brain ROI segmentation of multimodal tumor brain images. Due to the utilization of multimodal information and the reduced influence from pathological regions, experimental results show that registration based on our method is more accurate and robust, leading to significantly improved brain ROI segmentation compared with state-of-the-art methods.

Patch spaces and fusion strategies in patch-based label fusion

In the field of multi-atlas segmentation, patch-based approaches have shown promising results in the segmentation of biomedical images. In the most common approach, registration is used to warp the atlases to the target space and then the warped atlas labelmaps are fused into a consensus segmentation based on local appearance information encoded in form of patches. The registration step establishes spatial correspondence, which is important to obtain anatomical priors. Patch-based label fusion in the target space has shown to produce very accurate segmentations although at the expense of registering all atlases to each target image. Moreover, appearance (i.e., patches) and label information used by label fusion is extracted from the warped atlases, which are subject to interpolation errors. In this work, we revisit and extend the patch-based label fusion framework, exploring the role of extracting this information from the native space of both atlases and target images, thus avoiding interpolation artifacts, but at the same time, we do it in a way that it does not sacrifice the anatomical priors derived by registration. We further propose a common formulation for two widely-used label fusion strategies, i.e., similarity-based and a particular type of learning-based label fusion. The proposed framework is evaluated on subcortical structure segmentation in adult brains and tissue segmentation in fetal brain MRI. Our results indicate that using atlas patches in their native space yields superior performance than warping the atlases to the target image. The learning-based approach tends to outperform the similarity-based approach, with the particularity that using patches in native space lessens the computational requirements of learning. As conclusion, the combination of learning-based label fusion and native atlas patches yields the best performance with reduced test times than conventional similarity-based approaches.

Comparing fully automated state-of-the-art cerebellum parcellation from magnetic resonance images

The human cerebellum plays an essential role in motor control, is involved in cognitive function (i.e., attention, working memory, and language), and helps to regulate emotional responses. Quantitative in-vivo assessment of the cerebellum is important in the study of several neurological diseases including cerebellar ataxia, autism, and schizophrenia. Different structural subdivisions of the cerebellum have been shown to correlate with differing pathologies. To further understand these pathologies, it is helpful to automatically parcellate the cerebellum at the highest fidelity possible. In this paper, we coordinated with colleagues around the world to evaluate automated cerebellum parcellation algorithms on two clinical cohorts showing that the cerebellum can be parcellated to a high accuracy by newer methods. We characterize these various methods at four hierarchical levels: coarse (i.e., whole cerebellum and gross structures), lobe, subdivisions of the vermis, and the lobules. Due to the number of labels, the hierarchy of labels, the number of algorithms, and the two cohorts, we have restricted our analyses to the Dice measure of overlap. Under these conditions, machine learning based methods provide a collection of strategies that are efficient and deliver parcellations of a high standard across both cohorts, surpassing previous work in the area. In conjunction with the rank-sum computation, we identified an overall winning method.

Task-enhanced arterial spin labeled perfusion MRI predicts longitudinal neurodegeneration in mild cognitive impairment

Mild cognitive impairment (MCI) is considered a prodromal stage of Alzheimer's disease (AD), but is also recognized to be a heterogeneous condition. Biomarkers that predict AD progression in MCI are of clinical significance because they can be used to better identify appropriate candidates for therapeutic intervention studies. It has been hypothesized that comparing to structural measurements, functional ones may be more sensitive to early disease abnormalities and the sensitivity could be further enhanced when combined with cognitive task, a "brain stress test." In this study, we investigated the value of regional cerebral blood flow (CBF), measured by arterial spin labeled perfusion MRI (ASL MRI) during a memory-encoding task, in predicting the estimated rate of hippocampal atrophy, an established marker of AD progression. Thirty-one amnestic MCI patients (20 male and 11 female; age: 70.9 ± 6.5 years, range from 56 to 83 years; mini mental status examination: 27.8 ± 1.8) and 42 normal control subjects (13 male and 29 female; age: 70.6 ± 8.8 years, range from 55 to 88 years; mini mental status examination: 29.1 ± 1.2) were included in this study. We compared the predictive value of CBF during task to CBF during rest and structural volumetry. Both region-of-interest and voxelwise analyses showed that baseline CBF measurements during task (strongest effect in fusiform gyrus, region-of-interest analysis statistics: r = 0.56, p = .003), but not resting ASL MRI or structural volumetry, were correlated with the estimated rate of hippocampal atrophy in amnestic MCI patients. Further, stepwise linear regression demonstrated that resting ASL MRI and volumetry did not provide complementary information in prediction. These results support the notion that physiologic measures during a cognitive challenge may increase the ability to detect subtle functional changes that predict progression. As such, ASL MRI could have important utility in stratifying candidates for AD treatment trials.

The Role of Language Severity and Education in Explaining Performance on Object and Action Naming in Primary Progressive Aphasia

Despite the common assumption that atrophy in a certain brain area would compromise the function that it subserves, this is not always the case, especially in complex clinical syndromes such as primary progressive aphasia (PPA). Clinical and demographic information may contribute to PPA phenotypes and explain the manifested impairments better than atrophy. In the present study, we asked how much variance of the object and action naming impairments observed in PPA may be attributed to atrophy in the language network alone vs. additional clinical and demographic factors including language severity and education. Thirty-nine participants with PPA underwent magnetic resonance imaging (MRI) for volumetric analysis and a complete neuropsychological examination, including standardized tests of object and action naming. We used stepwise regression models to compare atrophy (volumetric model) to clinical/demographic variables (clinical-demographic model) for naming objects and actions. The clinical-demographic model was the best-fit model that explained the largest amount of variance in both object and action naming. Brain volume measurements alone explained little variance in both object and action naming. Clinical factors, particularly language severity, and demographic factors, particularly education, need to be considered in conjunction with brain volumes in PPA. The present study emphasizes the complexity of PPA as a syndrome and provides an example of how volumetric, clinical and demographic factors may interact in determining naming performance/deterioration.

Shortcomings of Ventricle Segmentation Using Deep Convolutional Networks

Normal Pressure Hydrocephalus (NPH) is a brain disorder that can present with ventriculomegaly and dementia-like symptoms, which often can be reversed through surgery. Having accurate segmentation of the ventricular system into its sub-compartments from magnetic resonance images (MRI) would be beneficial to better characterize the condition of NPH patients. Previous segmentation algorithms need long processing time and often fail to accurately segment severely enlarged ventricles in NPH patients. Recently, deep convolutional neural network (CNN) methods have been reported to have fast and accurate performance on medical image segmentation tasks. In this paper, we present a 3D U-net CNN-based network to segment the ventricular system in MRI. We trained three networks on different data sets and compared their performances. The networks trained on healthy controls (HC) failed in patients with NPH pathology, even in patients with normal appearing ventricles. The network trained on images from HC and NPH patients provided superior performance against state-of-the-art methods when evaluated on images from both data sets.

A comparison of automated segmentation and manual tracing in estimating hippocampal volume in ischemic stroke and healthy control participants

Manual quantification of the hippocampal atrophy state and rate is time consuming and prone to poor reproducibility, even when performed by neuroanatomical experts. The automation of hippocampal segmentation has been investigated in normal aging, epilepsy, and in Alzheimer's disease. Our first goal was to compare manual and automated hippocampal segmentation in ischemic stroke and to, secondly, study the impact of stroke lesion presence on hippocampal volume estimation. We used eight automated methods to segment T1-weighted MR images from 105 ischemic stroke patients and 39 age-matched controls sampled from the Cognition And Neocortical Volume After Stroke (CANVAS) study. The methods were: AdaBoost, Atlas-based Hippocampal Segmentation (ABHS) from the IDeALab, Computational Anatomy Toolbox (CAT) using 3 atlas variants (Hammers, LPBA40 and Neuromorphometics), FIRST, FreeSurfer v5.3, and FreeSurfer v6.0-Subfields. A number of these methods were employed to re-segment the T1 images for the stroke group after the stroke lesions were masked (i.e., removed). The automated methods were assessed on eight measures: process yield (i.e. segmentation success rate), correlation (Pearson's R and Shrout's ICC), concordance (Lin's RC and Kandall's W), slope 'a' of best-fit line from correlation plots, percentage of outliers from Bland-Altman plots, and significance of control-stroke difference. We eliminated the redundant measures after analysing between-measure correlations using Spearman's rank correlation. We ranked the automated methods based on the sum of the remaining non-redundant measures where each measure ranged between 0 and 1. Subfields attained an overall score of 96.3%, followed by AdaBoost (95.0%) and FIRST (94.7%). CAT using the LPBA40 atlas inflated hippocampal volumes the most, while the Hammers atlas returned the smallest volumes overall. FIRST (p = 0.014), FreeSurfer v5.3 (p = 0.007), manual tracing (p = 0.049), and CAT using the Neuromorphometics atlas (p = 0.017) all showed a significantly reduced hippocampal volume mean for the stroke group compared to control at three months. Moreover, masking of the stroke lesions prior to segmentation resulted in hippocampal volumes which agreed less with manual tracing. These findings recommend an automated segmentation without lesion masking as a more reliable procedure for the estimation of hippocampal volume in ischemic stroke.

Integrating Semi-supervised and Supervised Learning Methods for Label Fusion in Multi-Atlas Based Image Segmentation

A novel label fusion method for multi-atlas based image segmentation method is developed by integrating semi-supervised and supervised machine learning techniques. Particularly, our method is developed in a pattern recognition based multi-atlas label fusion framework. We build random forests classification models for each image voxel to be segmented based on its corresponding image patches of atlas images that have been registered to the image to be segmented. The voxelwise random forests classification models are then applied to the image to be segmented to obtain a probabilistic segmentation map. Finally, a semi-supervised label propagation method is adapted to refine the probabilistic segmentation map by propagating its reliable voxelwise segmentation labels, taking into consideration consistency of local and global image appearance of the image to be segmented. The proposed method has been evaluated for segmenting the hippocampus in MR images and compared with alternative machine learning based multi-atlas based image segmentation methods. The experiment results have demonstrated that our method could obtain competitive segmentation performance (average Dice index > 0.88), compared with alternative multi-atlas based image segmentation methods under comparison. Source codes of the methods under comparison are publicly available at www.nitrc.org/frs/?group_id=1242.

Multi-Atlas Segmentation of MR Tumor Brain Images Using Low-Rank Based Image Recovery

We introduce a new multi-atlas segmentation (MAS) framework for MR tumor brain images. The basic idea of MAS is to register and fuse label information from multiple normal brain atlases to a new brain image for segmentation. Many MAS methods have been proposed with success. However, most of them are developed for normal brain images, and tumor brain images usually pose a great challenge for them. This is because tumors cause difficulties in registration of normal brain atlases to the tumor brain image. To address this challenge, in the first step of our MAS framework, a new low-rank method is used to get the recovered image of normal-looking brain from the MR tumor brain image based on the information of normal brain atlases. Different from conventional low-rank methods that produce the recovered image with distorted normal brain regions, our low-rank method harnesses a spatial constraint to get the recovered image with preserved normal brain regions. Then in the second step, normal brain atlases can be registered to the recovered image without influence from tumors. These two steps are iteratively proceeded until convergence, for obtaining the final segmentation of the tumor brain image. During the iteration, both the recovered image and the registration of normal brain atlases to the recovered image are gradually refined. We have compared our proposed method with state-of-the-art methods by using both synthetic and real MR tumor brain images. Experimental results show that our proposed method can get effectively recovered images and also improves segmentation accuracy.

Autosegmentation for thoracic radiation treatment planning: A grand challenge at AAPM 2017

PURPOSE

This report presents the methods and results of the Thoracic Auto-Segmentation Challenge organized at the 2017 Annual Meeting of American Association of Physicists in Medicine. The purpose of the challenge was to provide a benchmark dataset and platform for evaluating performance of autosegmentation methods of organs at risk (OARs) in thoracic CT images.

METHODS

Sixty thoracic CT scans provided by three different institutions were separated into 36 training, 12 offline testing, and 12 online testing scans. Eleven participants completed the offline challenge, and seven completed the online challenge. The OARs were left and right lungs, heart, esophagus, and spinal cord. Clinical contours used for treatment planning were quality checked and edited to adhere to the RTOG 1106 contouring guidelines. Algorithms were evaluated using the Dice coefficient, Hausdorff distance, and mean surface distance. A consolidated score was computed by normalizing the metrics against interrater variability and averaging over all patients and structures.

RESULTS

The interrater study revealed highest variability in Dice for the esophagus and spinal cord, and in surface distances for lungs and heart. Five out of seven algorithms that participated in the online challenge employed deep-learning methods. Although the top three participants using deep learning produced the best segmentation for all structures, there was no significant difference in the performance among them. The fourth place participant used a multi-atlas-based approach. The highest Dice scores were produced for lungs, with averages ranging from 0.95 to 0.98, while the lowest Dice scores were produced for esophagus, with a range of 0.55-0.72.

CONCLUSION

The results of the challenge showed that the lungs and heart can be segmented fairly accurately by various algorithms, while deep-learning methods performed better on the esophagus. Our dataset together with the manual contours for all training cases continues to be available publicly as an ongoing benchmarking resource.

An iterative multi-atlas patch-based approach for cortex segmentation from neonatal MRI

Brain structure analysis in the newborn is a major health issue. This is especially the case for preterm neonates, in order to obtain predictive information related to the child development. In particular, the cortex is a structure of interest, that can be observed in magnetic resonance imaging (MRI). However, neonatal MRI data present specific properties that make them challenging to process. In this context, multi-atlas approaches constitute an efficient strategy, taking advantage of images processed beforehand. The method proposed in this article relies on such a multi-atlas strategy. More precisely, it uses two paradigms: first, a non-local model based on patches; second, an iterative optimization scheme. Coupling both concepts allows us to consider patches related not only to the image information, but also to the current segmentation. This strategy is compared to other multi-atlas methods proposed in the literature. Experiments on dHCP datasets show that the proposed approach provides robust cortex segmentation results.

Automatic Segmentation of Corpus Callosum in Midsagittal Based on Bayesian Inference Consisting of Sparse Representation Error and Multi-Atlas Voting

In this paper, we introduce a novel automatic method for Corpus Callosum (CC) in midsagittal plane segmentation. The robust segmentation of CC in midsagittal plane is key role for quantitative study of structural features of CC associated with various neurological disorder such as epilepsy, autism, Alzheimer's disease, and so on. Our approach is based on Bayesian inference using sparse representation and multi-atlas voting which both methods are used in various medical imaging, and show outstanding performance. Prior information in the proposed Bayesian inference is obtained from probability map generated from multi-atlas voting. The probability map contains the information of shape and location of CC of target image. Likelihood in the proposed Bayesian inference is obtained from gamma distribution function, generated from reconstruction errors (or sparse representation error), which are calculated in sparse representation of target patch using foreground dictionary and background dictionary each. Unlike the usual sparse representation method, we added gradient magnitude and gradient direction information to the patches of dictionaries and target, which had better segmentation performance than when not added. We compared three main segmentation results as follow: (1) the joint label fusion (JLF) method which is state-of-art method in multi-atlas voting based segmentation for evaluation of our method; (2) prior information estimated from multi-atlas voting only; (3) likelihood estimated from comparison of the reconstruction errors from sparse representation error only; (4) the proposed Bayesian inference. The methods were evaluated using two data sets of T1-weighted images, which one data set consists of 100 normal young subjects and the other data set consist of 25 normal old subjects and 22 old subjects with heavy drinker. In both data sets, the proposed Bayesian inference method has significantly the best segmentation performance than using each method separately.

Brainstem enlargement in preschool children with autism: Results from an intermethod agreement study of segmentation algorithms

The intermethod agreement between automated algorithms for brainstem segmentation is investigated, focusing on the potential involvement of this structure in Autism Spectrum Disorders (ASD). Inconsistencies highlighted in previous studies on brainstem in the population with ASD may in part be a result of poor agreement in the extraction of structural features between different methods. A sample of 76 children with ASD and 76 age-, gender-, and intelligence-matched controls was considered. Volumetric analyses were performed using common tools for brain structures segmentation, namely FSL-FIRST, FreeSurfer (FS), and Advanced Normalization Tools (ANTs). For shape analysis SPHARM-MAT was employed. Intermethod agreement was quantified in terms of Pearson correlations between pairs of volumes obtained by the different methods. The degree of overlap between segmented masks was quantified in terms of the Dice index. Both Pearson correlations and Dice indices, showed poor agreement between FSL-FIRST and the other methods (ANTs and FS), which by contrast, yielded Pearson correlations greater than 0.93 and average Dice indices greater than 0.76 when compared with each other. As with volume, shape analyses exhibited discrepancies between segmentation methods, with particular differences noted between FSL-FIRST and the others (ANT and FS), with under- and over-segmentation in specific brainstem regions. These data suggest that research on brain structure alterations should cross-validate findings across multiple methods. We consistently detected an enlargement of brainstem volume in the whole sample and in the male cohort across multiple segmentation methods, a feature particularly driven by the subgroup of children with idiopathic intellectual disability associated with ASD.

Learning to combine complementary segmentation methods for fetal and 6-month infant brain MRI segmentation

Segmentation of brain structures during the pre-natal and early post-natal periods is the first step for subsequent analysis of brain development. Segmentation techniques can be roughly divided into two families. The first, which we denote as registration-based techniques, rely on initial estimates derived by registration to one (or several) templates. The second family, denoted as learning-based techniques, relate imaging (and spatial) features to their corresponding anatomical labels. Each approach has its own qualities and both are complementary to each other. In this paper, we explore two ensembling strategies, namely, stacking and cascading to combine the strengths of both families. We present experiments on segmentation of 6-month infant brains and a cohort of fetuses with isolated non-severe ventriculomegaly (INSVM). INSVM is diagnosed when ventricles are mildly enlarged and no other anomalies are apparent. Prognosis is difficult based solely on the degree of ventricular enlargement. In order to find markers for a more reliable prognosis, we use the resulting segmentations to find abnormalities in the cortical folding of INSVM fetuses. Segmentation results show that either combination strategy outperform all of the individual methods, thus demonstrating the capability of learning systematic combinations that lead to an overall improvement. In particular, the cascading strategy outperforms the ensembling one, the former one obtaining top 5, 7 and 13 results (out of 21 teams) in the segmentation of white matter, gray matter and cerebro-spinal fluid in the iSeg2017 MICCAI Segmentation Challenge. The resulting segmentations reveal that INSVM fetuses have a less convoluted cortex. This points to cortical folding abnormalities as potential markers of later neurodevelopmental outcomes.