Automatic Segmentation Of Brain Structures Research Articles

EA-Net: Edge-aware network for brain structure segmentation via decoupled high and low frequency features

Automatic brain structure segmentation in Magnetic Resonance Image (MRI) plays an important role in the diagnosis of various neuropsychiatric diseases. However, most existing methods yield unsatisfactory results due to blurred boundaries and complex structures. Improving the segmentation ability requires the model to be explicit about the spatial localization and shape appearance of targets, which correspond to the low-frequency content features and the high-frequency edge features, respectively. Therefore, in this paper, to extract rich edge and content feature representations, we focus on the composition of the feature and utilize a frequency decoupling (FD) block to separate the low-frequency and high-frequency parts of the feature. Further, a novel edge-aware network (EA-Net) is proposed for jointly learning to segment brain structures and detect object edges. First, an encoder–decoder sub-network is utilized to obtain multi-level information from the input MRI, which is then sent to the FD block to complete the frequency separation. Further, we use different mechanisms to optimize both the low-frequency and high-frequency features. Finally, these two parts are fused to generate the final prediction. In particular, we extract the content mask and the edge mask from the optimization feature with different supervisions, which forces the network to learn the boundary features of the object. Extensive experiments are performed on two public brain MRI T1 scan datasets (the IBSR dataset and the MALC dataset) to evaluate the effectiveness of the proposed algorithm. The experiments show that the EA-Net achieves the best performance compared with the state-of-the-art methods, and improves the segmentation DSC score by 1.31% at most compared with the U-Net model and its variants. Moreover, we evaluate the EA-Net under different noise disturbances, and the results demonstrate the robustness and superiority of our method under low-quality noisy MRI. Code is available at https://github.com/huqian999/EA-Net.

MO-0885 Automatic segmentation of brain structures in longitudinal MR images of growing children

MO-0885 Automatic segmentation of brain structures in longitudinal MR images of growing children

PO-1626 Automatic brain structure segmentation in children with brain tumours

PO-1626 Automatic brain structure segmentation in children with brain tumours

Validation of an automatic tool for the rapid measurement of brain atrophy and white matter hyperintensity: QyScore\xae

ObjectivesQyScore® is an imaging analysis tool certified in Europe (CE marked) and the US (FDA cleared) for the automatic volumetry of grey and white matter (GM and WM respectively), hippocampus (HP), amygdala (AM), and white matter hyperintensity (WMH). Here we compare QyScore® performances with the consensus of expert neuroradiologists.MethodsDice similarity coefficient (DSC) and the relative volume difference (RVD) for GM, WM volumes were calculated on 50 3DT1 images. DSC and the F1 metrics were calculated for WMH on 130 3DT1 and FLAIR images. For each index, we identified thresholds of reliability based on current literature review results. We hypothesized that DSC/F1 scores obtained using QyScore® markers would be higher than the threshold. In contrast, RVD scores would be lower. Regression analysis and Bland–Altman plots were obtained to evaluate QyScore® performance in comparison to the consensus of three expert neuroradiologists.ResultsThe lower bound of the DSC/F1 confidence intervals was higher than the threshold for the GM, WM, HP, AM, and WMH, and the higher bounds of the RVD confidence interval were below the threshold for the WM, GM, HP, and AM. QyScore®, compared with the consensus of three expert neuroradiologists, provides reliable performance for the automatic segmentation of the GM and WM volumes, and HP and AM volumes, as well as WMH volumes.ConclusionsQyScore® represents a reliable medical device in comparison with the consensus of expert neuroradiologists. Therefore, QyScore® could be implemented in clinical trials and clinical routine to support the diagnosis and longitudinal monitoring of neurological diseases.Key Points• QyScore® provides reliable automatic segmentation of brain structures in comparison with the consensus of three expert neuroradiologists.• QyScore® automatic segmentation could be performed on MRI images using different vendors and protocols of acquisition. In addition, the fast segmentation process saves time over manual and semi-automatic methods.• QyScore® could be implemented in clinical trials and clinical routine to support the diagnosis and longitudinal monitoring of neurological diseases.

Association between symbol digit modalities test and regional cortex thickness in young adults with relapsing-remitting multiple sclerosis

BackgroundMultiple sclerosis (MS) is a demyelinating disease of the central nervous system, predominating within young adults. Cognitive disorders are common in MS and have are associated with several Magnetic Resonance Imaging (MRI) markers, especially brain atrophy. Many have found the symbol digit modalities test (SDMT) to be the most sensitive individual cognitive measure relevant to MS. However, the relationship between SDMT and regional brain cortex thickness in young adults with relapsing-remitting multiple sclerosis (YA-RRMS) has been little explored. The purpose of this study was to investigate the association between the SDMT and regional cortex thickness in YA-RRMS by FreeSurfer, which is an automatic brain structure segmentation method. MethodTwenty-eight YA-RRMS patients (18–35 years old) were enrolled in the present study. Informed consent and information including gender, age, disease duration, number of relapses, annual relapse rate was collected from all patients. Clinical cognitive evaluations (SDMT and auditory verbal learning test (AVLT)) and daily performance: activities of daily living (ADL) were assessed in the present study. MRI scans were performed at the Institute of Neurosurgery of Tiantan Hospital. Twenty-eight matched healthy controls (HC) MRI data were obtained from Tiantan Hospital database. Data on thirty-four points of bilateral cortical structure thickness using statistically defined brain regions-of-interest from FreeSurfer were obtained from all participants. ResultsPatients with RRMS exhibited extensively thinner cerebellar cortex compared with HC. SDMT scores were significantly correlated with AVLT subentries (IM, immediate memory; DRM, delayed recall memory; LTRM, long-term recognition memory) in YA-RRMS patients (P < 0.05). SDMT was strongly correlated with regional cortex thickness differences of the right temporal pole (r = 0.68) and bilateral parahippocampal areas (right r = 0.62; left r = 0.60), and moderately correlated with regional cortex thickness differences including the left superior temporal and right insula (r = 0.57 and 0.56, respectively) in YA-RRMS patients. ConclusionThe present study has shown the SDMT is strongly correlated with selected cortex regions including the bilateral parahippocampal area and the right temporal pole which are involved in geometric structures processing.

Automatic segmentation of brain MRI using a novel patch-wise U-net deep architecture.

Accurate segmentation of brain magnetic resonance imaging (MRI) is an essential step in quantifying the changes in brain structure. Deep learning in recent years has been extensively used for brain image segmentation with highly promising performance. In particular, the U-net architecture has been widely used for segmentation in various biomedical related fields. In this paper, we propose a patch-wise U-net architecture for the automatic segmentation of brain structures in structural MRI. In the proposed brain segmentation method, the non-overlapping patch-wise U-net is used to overcome the drawbacks of conventional U-net with more retention of local information. In our proposed method, the slices from an MRI scan are divided into non-overlapping patches that are fed into the U-net model along with their corresponding patches of ground truth so as to train the network. The experimental results show that the proposed patch-wise U-net model achieves a Dice similarity coefficient (DSC) score of 0.93 in average and outperforms the conventional U-net and the SegNet-based methods by 3% and 10%, respectively, for on Open Access Series of Imaging Studies (OASIS) and Internet Brain Segmentation Repository (IBSR) dataset.

Brain Structures Segmentation by using Statistical Models

Automatic segmentation of brain structures is a fundamental step for quantitative analysis of images in many brain’s pathologies such as Alzheimer’s, brain’s tumors or multiple sclerosis. The large variation of brain structures requires the development of efficient and specific methods, often by using Magnetic Resonance Imaging (MRI) modality. The goal of our work is to implement an automatic brain’s structures segmentation method that uses the active shape models (ASM) and active appearance models (AAM) techniques. Another goal of this work is to compare the performances of these segmentation approaches, and also to evaluate their use in a computer aided diagnosis tools and to compare their performances.

Evaluating the effect of multiple sclerosis lesions on automatic brain structure segmentation.

In recent years, many automatic brain structure segmentation methods have been proposed. However, these methods are commonly tested with non-lesioned brains and the effect of lesions on their performance has not been evaluated. Here, we analyze the effect of multiple sclerosis (MS) lesions on three well-known automatic brain structure segmentation methods, namely, FreeSurfer, FIRST and multi-atlas fused by majority voting, which use learning-based, deformable and atlas-based strategies, respectively. To perform a quantitative analysis, 100 synthetic images of MS patients with a total of 2174 lesions are simulated on two public databases with available brain structure ground truth information (IBSR18 and MICCAI’12). The Dice similarity coefficient (DSC) differences and the volume differences between the healthy and the simulated images are calculated for the subcortical structures and the brainstem. We observe that the three strategies are affected when lesions are present. However, the effects of the lesions do not follow the same pattern; the lesions either make the segmentation method underperform or surprisingly augment the segmentation accuracy. The obtained results show that FreeSurfer is the method most affected by the presence of lesions, with DSC differences (generated − healthy) ranging from − 0.11 ± 0.54 to 9.65 ± 9.87, whereas FIRST tends to be the most robust method when lesions are present (− 2.40 ± 5.54 to 0.44 ± 0.94). Lesion location is not important for global strategies such as FreeSurfer or majority voting, where structure segmentation is affected wherever the lesions exist. On the other hand, FIRST is more affected when the lesions are overlaid or close to the structure of analysis. The most affected structure by the presence of lesions is the nucleus accumbens (from − 1.12 ± 2.53 to 1.32 ± 4.00 for the left hemisphere and from − 2.40 ± 5.54 to 9.65 ± 9.87 for the right hemisphere), whereas the structures that show less variation include the thalamus (from 0.03 ± 0.35 to 0.74 ± 0.89 and from − 0.48 ± 1.08 to − 0.04 ± 0.22) and the brainstem (from − 0.20 ± 0.38 to 1.03 ± 1.31). The three segmentation approaches are affected by the presence of MS lesions, which demonstrates that there exists a problem in the automatic segmentation methods of the deep gray matter (DGM) structures that has to be taken into account when using them as a tool to measure the disease progression.

Clinical application of apparent diffusion coefficient mapping in voxel-based morphometry in the diagnosis of Alzheimer's disease

To evaluate the performance of apparent diffusion coefficient (ADC) mapping compared with voxel-based morphometry and to demonstrate the clinical usefulness of ADC mapping in the diagnosis of Alzheimer's disease (AD). The study population comprised 31 patients with AD (group A) and 24 patients without dementia (group B) who underwent three-dimensional (3D) T1-weighted imaging (WI) and two-dimensional (2D) echo-planar diffusion-weighted imaging (DWI) at 3T. The volume and ADC of the regional grey matter (GM) in the bilateral hippocampi, precunei, and the anterior and posterior cingulate gyri were calculated using a voxel-based method for automatic segmentation of brain structures. The significance of intergroup differences in each volume and ADC of all regional GM was tested using analysis of variance (ANOVA) with Bonferroni correction. Intergroup regional GM differences in each volume and ADC were evaluated using statistical parametric mapping (SPM). In group A, the volumes of the precunei (mean value: group A/B=18.93/21.48cm3) and the anterior cingulate gyri (mean value: group A/B=6.1/7.81cm3) were significantly less than in group B (p<0.05). The ADC in group A was significantly larger than that in group B in the bilateral hippocampi (mean value: group A/B=right 1020.79×10-6/877.23×10-6mm2/s; left 1072.89×10-6/900.2×10-6mm2/s) and posterior cingulate gyri (mean value: group A/B=1006.77×10-6/876.88×10-6mm2/s; p<0.05). SPM showed that the areas of increased ADC were more extensive than the areas of decreased volume in the bilateral hippocampi, precunei, and posterior cingulate gyri in group A, compared with those in group B (p<0.001). Evaluation of ADC mapping can quantify changes in brain water diffusivity and may improve the performance of automatic morphometric diagnosis of AD.

Hippocampus and amygdala volumes from magnetic resonance images in children: Assessing accuracy of FreeSurfer and FSL against manual segmentation

The volumetric quantification of brain structures is of great interest in pediatric populations because it allows the investigation of different factors influencing neurodevelopment. FreeSurfer and FSL both provide frequently used packages for automatic segmentation of brain structures. In this study, we examined the accuracy and consistency of those two automated protocols relative to manual segmentation, commonly considered as the “gold standard” technique, for estimating hippocampus and amygdala volumes in a sample of preadolescent children aged between 6 to 11years. The volumes obtained with FreeSurfer and FSL-FIRST were evaluated and compared with manual segmentations with respect to volume difference, spatial agreement and between- and within-method correlations.Results highlighted a tendency for both automated techniques to overestimate hippocampus and amygdala volumes, in comparison to manual segmentation. This was more pronounced when using FreeSurfer than FSL-FIRST and, for both techniques, the overestimation was more marked for the amygdala than the hippocampus. Pearson correlations support moderate associations between manual tracing and FreeSurfer for hippocampus (right r=0.69, p<0.001; left r=0.77, p<0.001) and amygdala (right r=0.61, p<0.001; left r=0.67, p<0.001) volumes. Correlation coefficients between manual segmentation and FSL-FIRST were statistically significant (right hippocampus r=0.59, p<0.001; left hippocampus r=0.51, p<0.001; right amygdala r=0.35, p<0.001; left amygdala r=0.31, p<0.001) but were significantly weaker, for all investigated structures. When computing intraclass correlation coefficients between manual tracing and automatic segmentation, all comparisons, except for left hippocampus volume estimated with FreeSurfer, failed to reach 0.70. When looking at each method separately, correlations between left and right hemispheric volumes showed strong associations between bilateral hippocampus and bilateral amygdala volumes when assessed using manual segmentation or FreeSurfer. These correlations were significantly weaker when volumes were assessed with FSL-FIRST. Finally, Bland–Altman plots suggest that the difference between manual and automatic segmentation might be influenced by the volume of the structure, because smaller volumes were associated with larger volume differences between techniques.These results demonstrate that, at least in a pediatric population, the agreement between amygdala and hippocampus volumes obtained with automated FSL-FIRST and FreeSurfer protocols and those obtained with manual segmentation is not strong. Visual inspection by an informed individual and, if necessary, manual correction of automated segmentation outputs are important to ensure validity of volumetric results and interpretation of related findings.

Statistical validation of automatic methods for hippocampus segmentation in MR images of epileptic patients.

Hippocampus segmentation is a key step in the evaluation of mesial Temporal Lobe Epilepsy (mTLE) by MR images. Several automated segmentation methods have been introduced for medical image segmentation. Because of multiple edges, missing boundaries, and shape changing along its longitudinal axis, manual outlining still remains the benchmark for hippocampus segmentation, which however, is impractical for large datasets due to time constraints. In this study, four automatic methods, namely FreeSurfer, Hammer, Automatic Brain Structure Segmentation (ABSS), and LocalInfo segmentation, are evaluated to find the most accurate and applicable method that resembles the bench-mark of hippocampus. Results from these four methods are compared against those obtained using manual segmentation for T1-weighted images of 157 symptomatic mTLE patients. For performance evaluation of automatic segmentation, Dice coefficient, Hausdorff distance, Precision, and Root Mean Square (RMS) distance are extracted and compared. Among these four automated methods, ABSS generates the most accurate results and the reproducibility is more similar to expert manual outlining by statistical validation. By considering p-value<;0.05, the results of performance measurement for ABSS reveal that, Dice is 4%, 13%, and 17% higher, Hausdorff is 23%, 87%, and 70% lower, precision is 5%, -5%, and 12% higher, and RMS is 19%, 62%, and 65% lower compared to LocalInfo, FreeSurfer, and Hammer, respectively.