Figure - available from: Human Brain Mapping
This content is subject to copyright. Terms and conditions apply.
Dentate, red nucleus, substantia nigra, subthalamic nuclei visible as hypointense regions in a normalized SW image (a,b), and their corresponding labels (d,e). The location of the axial views is indicated by the blue lines in the sagittal view (c): top (a, d) and bottom (b, e)
Source publication
The advent of susceptibility-sensitive MRI techniques, such as susceptibility weighted imaging (SWI), has enabled accurate in vivo visualization and quantification of iron deposition within the human brain. Although previous approaches have been introduced to segment iron-rich brain regions, such as the substantia nigra, sub-thalamic nucleus, red n...
Citations
... HiRes segmentation of brain structures has mostly been tackled by training with manual annotations created at the desired resolution (Beliveau et al., 2021;Estrada et al., 2021;Kamnitsas et al., 2017;Rushmore et al., 2022) or training models using 1.0 mm data with scale-augmentations -an established deeplearning technique to improve the generalizability of a model. Recently, models capable of segmenting scans at different resolutions have been introduced. ...
The hypothalamus plays a crucial role in the regulation of a broad range of physiological, behavioural, and cognitive functions. However, despite its importance, only a few small-scale neuroimaging studies have investigated its substructures, likely due to the lack of fully automated segmentation tools to address scalability and reproducibility issues of manual segmentation. While the only previous attempt to automatically sub-segment the hypothalamus with a neural network showed promise for 1.0 mm isotropic T1-weighted (T1w) MRI, there is a need for an automated tool to sub-segment also high-resolutional (HiRes) MR scans, as they are becoming widely available, and include structural detail also from multi-modal MRI. We, therefore, introduce a novel, fast, and fully automated deep learning method named HypVINN for sub-segmentation of the hypothalamus and adjacent structures on 0.8 mm isotropic T1w and T2w brain MR images that is robust to missing modalities. We extensively validate our model with respect to segmentation accuracy, generalizability, in-session test-retest reliability, and sensitivity to replicate hypothalamic volume effects (e.g. sex-differences). The proposed method exhibits high segmentation performance both for standalone T1w images as well as for T1w/T2w image pairs. Even with the additional capability to accept flexible inputs, our model matches or exceeds the performance of state-of-the-art methods with fixed inputs. We, further, demonstrate the generalizability of our method in experiments with 1.0 mm MR scans from both the Rhineland Study and the UK Biobank— an independent dataset never encountered during training with different acquisition parameters and demographics. Finally, HypVINN can perform the segmentation in less than a minute (GPU) and will be available in the open source FastSurfer neuroimaging software suite, offering a validated, efficient, and scalable solution for evaluating imaging-derived phenotypes of the hypothalamus.
... HiRes segmentation of brain structures has mostly been tackled by training with manual annotations created at the desired resolution [25,30,36,37] or training models using 1.0 mm data with scale-augmentations -an established deep-learning technique to improve the generalizability of a model. Recently, models capable of segmenting scans at different resolutions have been introduced. ...
The hypothalamus plays a crucial role in the regulation of a broad range of physiological, behavioural, and cognitive functions. However, despite its importance, only a few small-scale neuroimaging studies have investigated its substructures, likely due to the lack of fully automated segmentation tools to address scalability and reproducibility issues of manual segmentation. While the only previous attempt to automatically sub-segment the hypothalamus with a neural network showed promise for 1.0 mm isotropic T1-weighted (T1w) MRI, there is a need for an automated tool to sub-segment also high-resolutional (HiRes) MR scans, as they are becoming widely available, and include structural detail also from multi-modal MRI. We, therefore, introduce a novel, fast, and fully automated deep learning method named HypVINN for sub-segmentation of the hypothalamus and adjacent structures on 0.8 mm isotropic T1w and T2w brain MR images that is robust to missing modalities. We extensively validate our model with respect to segmentation accuracy, generalizability, in-session test-retest reliability, and sensitivity to replicate hypothalamic volume effects (e.g. sex-differences). The proposed method exhibits high segmentation performance both for standalone T1w images as well as for T1w/T2w image pairs. Even with the additional capability to accept flexible inputs, our model matches or exceeds the performance of state-of-the-art methods with fixed inputs. We, further, demonstrate the generalizability of our method in experiments with 1.0 mm MR scans from both the Rhineland Study and the UK Biobank. Finally, HypVINN can perform the segmentation in less than a minute (GPU) and will be available in the open source FastSurfer neuroimaging software suite, offering a validated, efficient, and scalable solution for evaluating imaging-derived phenotypes of the hypothalamus.
... As iron has been proned to accumulate in the gray matter nuclei in normal people and all these neurological diseases have abnormal iron deposition in the gray matter nuclei, the gray matter nuclei are the critical target structures to explore the abnormal iron deposition. Previous studies have found that routine structural MR images such as T 1 -weighted images could hardly show ironrich gray matter nuclei clearly, such as substantia nigra (SN), red nucleus (RN), and dentate nucleus (DN; Beliveau et al., 2021). Therefore, these nuclei were not found in the most popular brain atlas, including FreeSurfer, FMRIB Software Library (FSL), and Statistical Parametric Mapping (SPM). ...
... Therefore, these nuclei were not found in the most popular brain atlas, including FreeSurfer, FMRIB Software Library (FSL), and Statistical Parametric Mapping (SPM). Most segmentation tools cannot extract these nuclei (Beliveau et al., 2021). However, all the gray matter nuclei, including SN, RN, and DN, showed the obvious contrast (high signal) relative to the surrounding brain tissues in the QSM images because QSM was very sensitive to the iron, even when the amount was small and QSM can also enhance the iron-related contrast (Beliveau et al., 2021). ...
... Most segmentation tools cannot extract these nuclei (Beliveau et al., 2021). However, all the gray matter nuclei, including SN, RN, and DN, showed the obvious contrast (high signal) relative to the surrounding brain tissues in the QSM images because QSM was very sensitive to the iron, even when the amount was small and QSM can also enhance the iron-related contrast (Beliveau et al., 2021). The apparent contrast can help to identify the gray matter nuclei clearly and accurately. ...
The abnormal iron deposition of the deep gray matter nuclei is related to many neurological diseases. With the quantitative susceptibility mapping (QSM) technique, it is possible to quantitatively measure the brain iron content in vivo. To assess the magnetic susceptibility of the deep gray matter nuclei in the QSM, it is mandatory to segment the nuclei of interest first, and many automatic methods have been proposed in the literature. This study proposed a contrast attention U-Net for nuclei segmentation and evaluated its performance on two datasets acquired using different sequences with different parameters from different MRI devices. Experimental results revealed that our proposed method was superior on both datasets over other commonly adopted network structures. The impacts of training and inference strategies were also discussed, which showed that adopting test time augmentation during the inference stage can impose an obvious improvement. At the training stage, our results indicated that sufficient data augmentation, deep supervision, and nonuniform patch sampling contributed significantly to improving the segmentation accuracy, which indicated that appropriate choices of training and inference strategies were at least as important as designing more advanced network structures.
... Compared to PD, increased magnetic susceptibility was reported in the midbrain of PSP patients, mainly in the red nucleus (RN), and in the putamen of MSA patients [35]. The identification of diseasespecific susceptibility cut-off values could improve the diagnostic potential of QSM, in addition to newly available automated techniques for the segmentation of deep brain nuclei [36]. Although QSM has been widely applied in PD patients [37][38][39][40][41] and in different disease stages [40], a large variability of regional susceptibility values has been reported [42], possibly related to the heterogeneity of the methods applied across studies. ...
Background:
In the quest for in vivo diagnostic biomarkers to discriminate Parkinson's Disease (PD) from Progressive Supranuclear Palsy (PSP) and Multiple System Atrophy (MSA, mainly p phenotype), many advanced MRI techniques have been studied. Morphometric indexes, such as the Magnetic Resonance Parkinsonism Index (MRPI), demonstrated high diagnostic value in the comparison PD-PSP. The potential of Quantitative Susceptibility Mapping (QSM) was hypothesized, as increased magnetic susceptibility (Δχ) was reported in the red nucleus (RN) and medial part of substantia nigra (SNImed) of PSP patients, and in the putamen of MSA patients. However, disease-specific susceptibility values for relevant regions-of-interest are yet to be identified. Aims of the study were to evaluate the diagnostic potential of a multimodal MRI protocol combining morphometric and QSM imaging in patients with Determined parkinsonisms, and to explore its value in a population of Undetermined cases.
Method:
Patients with suspected degenerative parkinsonism underwent clinical evaluation, 3T brain MRI, and clinical follow-up. The MRPI was manually calculated on T1-weighted images. QSM maps were generated from 3D multi-echo T2*-weighted sequences.
Results:
In Determined cases the morphometric evaluation confirmed optimal diagnostic accuracy in the comparison PD-PSP but failed to discriminate PD from MSA-p. Significant nigral and extranigral differences were found with QSM. RN Δχ showed excellent diagnostic accuracy in the comparison PD-PSP and good accuracy in PD-MSA-p. Optimal susceptibility cut-off values of RN and SNImed were tested in Undetermined cases in addition to MRPI.
Conclusions:
A combined use of morphometric imaging and QSM could improve the diagnostic phase of degenerative parkinsonisms.
... CNN has also been used in segmenting sub-cortical brain structures in traditional T1-weighted MRI (Dolz et al., 2018). Deep learning approaches achieved better overall performance in automated segmentation of subcortical brain structures, compared with atlas-based approaches and algorithmic approaches (Pagnozzi et al., 2019;Beliveau et al., 2021). To overcome the issue that CNN-based methods require large training datasets, the transfer learning can be applied, where a network pre-trained with a much larger dataset is used to initialize the target network weights, significantly reducing the demand of training data and training time for the target network (Pan and Yang, 2009;Xu et al., 2017). ...
... A variety of midbrain structures segmentation studies have been reported (Lim et al., 2013;Li et al., 2016Li et al., , 2019Visser et al., 2016;Garzón et al., 2018;Kim et al., 2019;Plassard et al., 2019;Beliveau et al., 2021), which are summarized in Table 2. These studies were mainly based on multi-modality MRI images, including T1-, T2-, T2 * -weighted or QSM images. ...
... Convolutional neural network model with transfer learning is another attribution for our excellent segmentation performance. Up to now, no study has utilized CNN to segment midbrain structures in high-resolution susceptibility maps, though the CNN model is rapidly evolving in image segmentations (Kim et al., 2019;Beliveau et al., 2021). The size of the highresolution dataset in our study can hardly meet the requirement of CNN model training. ...
Background
Accurate delineation of the midbrain nuclei, the red nucleus (RN), substantia nigra (SN) and subthalamic nucleus (STN), is important in neuroimaging studies of neurodegenerative and other diseases. This study aims to segment midbrain structures in high-resolution susceptibility maps using a method based on a convolutional neural network (CNN).
Methods
The susceptibility maps of 75 subjects were acquired with a voxel size of 0.83 × 0.83 × 0.80 mm³ on a 3T MRI system to distinguish the RN, SN, and STN. A deeply supervised attention U-net was pre-trained with a dataset of 100 subjects containing susceptibility maps with a voxel size of 0.63 × 0.63 × 2.00 mm³ to provide initial weights for the target network. Five-fold cross-validation over the training cohort was used for all the models’ training and selection. The same test cohort was used for the final evaluation of all the models. Dice coefficients were used to assess spatial overlap agreement between manual delineations (ground truth) and automated segmentation. Volume and magnetic susceptibility values in the nuclei extracted with automated CNN delineation were compared to those extracted by manual tracing. Consistencies of volume and magnetic susceptibility values by different extraction strategies were assessed by Pearson correlation coefficients and Bland-Altman analyses.
Results
The automated CNN segmentation method achieved mean Dice scores of 0.903, 0.864, and 0.777 for the RN, SN, and STN, respectively. There were no significant differences between the achieved Dice scores and the inter-rater Dice scores (p > 0.05 for each nucleus). The overall volume and magnetic susceptibility values of the nuclei extracted by the automatic CNN method were significantly correlated with those by manual delineation (p < 0.01).
Conclusion
Midbrain structures can be precisely segmented in high-resolution susceptibility maps using a CNN-based method.
Purpose To develop a fast and fully automated deep learning (DL)-based method for the MRI planimetric segmentation and measurement of the brainstem and ventricular structures most affected in patients with progressive supranuclear palsy (PSP). Materials and Methods In this retrospective study, T1-weighted MR images in healthy controls (n = 84) were used to train DL models for segmenting the midbrain, pons, middle cerebellar peduncle (MCP), superior cerebellar peduncle (SCP), third ventricle, and frontal horns (FHs). Internal, external, and clinical test datasets (n = 305) were used to assess segmentation model reliability. DL masks from test datasets were used to automatically extract midbrain and pons areas and the width of MCP, SCP, third ventricle, and FHs. Automated measurements were compared with those manually performed by an expert radiologist. Finally, these measures were combined to calculate the midbrain to pons area ratio, MR parkinsonism index (MRPI), and MRPI 2.0, which were used to differentiate patients with PSP (n = 71) from those with Parkinson disease (PD) (n = 129). Results Dice coefficients above 0.85 were found for all brain regions when comparing manual and DL-based segmentations. A strong correlation was observed between automated and manual measurements (Spearman ρ > 0.80, P < .001). DL-based measurements showed excellent performance in differentiating patients with PSP from those with PD, with an area under the receiver operating characteristic curve above 0.92. Conclusion The automated approach successfully segmented and measured the brainstem and ventricular structures. DL-based models may represent a useful approach to support the diagnosis of PSP and potentially other conditions associated with brainstem and ventricular alterations. Keywords: MR Imaging, Brain/Brain Stem, Segmentation, Quantification, Diagnosis, Convolutional Neural Network Supplemental material is available for this article. © RSNA, 2024 See also the commentary by Mohajer in this issue.
Background:
Deep Brain Stimulation of the subthalamic nucleus (STN-DBS) is an established and effective neurosurgical treatment for relieving motor symptoms in Parkinson's disease (PD). The localization of key brain structures is critical to the success of DBS surgery. However, in clinical practice, this process is heavily dependent on the radiologists' experience.
Methods:
In this study, we propose an automatic localization method of key structures for STN-DBS surgery via prior-enhanced multi-object MRI segmentation. We utilize the U-Net architecture for the multi-object segmentation, including STN, red nucleus, brain sulci, gyri, and ventricles. To address the challenge that only half of the brain sulci and gyri locate in the upper area, potentially causing interference in the lower area, we perform region of interest (ROI) detection and ensemble joint processing to enhance the segmentation performance of brain sulci and gyri.
Results:
We evaluate the segmentation accuracy by comparing our method with other state-of-the-art machine learning segmentation methods. The experimental results demonstrate that our approach outperforms the current state-of-the-art methods in terms of segmentation performance. Moreover, our method provides effective visualization of key brain structures from clinical application perspective and can reduce the segmentation time when compared with manual delineation.
Conclusions:
Our proposed method utilizes deep learning to achieve accurate segmentation of the key structures more quickly and comparable accuracy with human manual segmentation. Our method has the potential to improve the efficiency of surgical planning for STN-DBS.
Background:
Predicting postoperative visual outcome of pituitary adenoma patients is important but remains challenging. This study aimed to identify a novel prognostic predictor which can be automatically obtained from routine MRI using deep learning approach.
Materials and methods:
A total of 220 pituitary adenoma patients were prospectively enrolled and stratified into the recovery and non-recovery group according to visual outcome at six months after endoscopic endonasal transsphenoidal surgery. The optic chiasm was manually segmented on preoperative coronal T2WI, and its morphometric parameters were measured including suprasellar extension distance, chiasmal thickness and chiasmal volume. Univariate and multivariate analyses were conducted on clinical and morphometric parameters to identify predictors for visual recovery. Additionally, a deep learning model for automated segmentation and volumetric measurement of optic chiasm was developed with nnU-Net architecture and evaluated in a multi-center dataset covering 1026 pituitary adenoma patients from 4 institutions.
Results:
Larger preoperative chiasmal volume was significantly associated with better visual outcome (P=0.001). Multivariate logistic regression suggested it could be taken as the independent predictor for visual recovery (odds ratio=2.838, P<0.001). The auto-segmentation model represented good performances and generalizability in internal (Dice=0.813) and three independent external test sets (Dice=0.786, 0.818 and 0.808, respectively). Moreover, the model achieved accurate volumetric evaluation of the optic chiasm with intraclass correlation coefficient of more than 0.83 in both internal and external test sets.
Conclusion:
The preoperative volume of the optic chiasm could be utilized as the prognostic predictor for visual recovery of pituitary adenoma patients after surgery. Moreover, the proposed deep learning-based model allowed for automated segmentation and volumetric measurement of the optic chiasm on routine MRI.
The establishment of an unbiased protocol for the automated volumetric measurement of iron‐rich regions in the substantia nigra (SN) is clinically important for diagnosing neurodegenerative diseases exhibiting midbrain atrophy, such as progressive supranuclear palsy (PSP). This study aimed to automatically quantify the volume and surface properties of the iron‐rich 3D regions in the SN using the quantitative MRI‐R2* map. 367 slices of R2* map and susceptibility‐weighted imaging (SWI) at 3T MRI from healthy control (HC) individuals and Parkinson’s disease (PD) patients were used to train customized U‐net++ convolutional neural network based on expert‐segmented masks. Age‐ and sex‐matched participants were selected from HC, PD, and PSP groups to automate the volumetric determination of iron‐rich areas in the SN. Dice similarity coefficient (DSC) values between expert‐segmented and detected masks from the proposed network were 0.91±0.07 for R2* maps and 0.89±0.08 for SWI. Reductions in iron‐rich SN volume from the R2* map (SWI) were observed in PSP with an area under the receiver operating characteristic curve values of 0.96 (0.89) and 0.98 (0.92) compared to HC and PD, respectively. The mean curvature of the PSP showed SN deformation along the side closer to the red nucleus. We demonstrated the automated volumetric measurement of iron‐rich regions in the SN using deep learning can quantify the SN atrophy in PSP compared to PD and HC.
Introduction:
Iron accumulation in subcortical brain nuclei has been shown to be differentially increased in atypical parkinsonian disorders. It is unclear whether the patterns of iron accumulation are consistent between variants of progressive supranuclear palsy (PSP) and their diagnostic utility in early to moderately advanced stages of the diseases.
Methods:
Brain iron content (R2*) was quantified using magnetic resonance imaging in patients clinically diagnosed as PSP - parkinsonism (PSP-P, n = 15), PSP - Richardson's syndrome (PSP-RS, n = 14), Parkinson's disease (PD, n = 15), or the parkinsonian variant of multiple system atrophy (MSA-P, n = 14) using established criteria, and healthy controls (HC). Disease duration was less than 5 years in all patients. The quantification of regional R2* was performed using a semi-automatized approach.
Results:
Significant group differences in R2*, primarily within the red nucleus and the substantia nigra, were identified between PSP, PD, MSA-P, and HC, but not between PSP-P and PSP-RS. However, distinct R2* correlation patterns across brain regions were observed for the different groups. Good classification performances (sensitivity and specificity >80%) were only obtained for PSP vs. HC. For all other comparisons, sensitivity and/or specificity was below <70%.
Conclusion:
Iron accumulation in subcortical brain nuclei has distinct correlated patterns in PSP-P and PSP-RS, which could be reflecting different pathophysiological mechanisms. Increased iron content in these nuclei appears to be a predictor for atypical parkinsonian disorders such as PSP and MSA. Further studies are required to reproduce this finding and elucidate the evolution of these patterns over the course of the disease.
