Figure 4 - available from: Scientific Reports
This content is subject to copyright. Terms and conditions apply.
This figure shows examples of cerebral microbleeds and basal ganglia iron deposition in SWI for TE = 7.5 ms (left column), SWI for TE = 22.5 ms (middle left column), QSM (middle right column) and T2w (right column). (A, C) Show the lesions in two different brains, and (B, D) show the corresponding human expert labeling of the CMBs (red) and iron deposits (green).
Source publication
Lobar cerebral microbleeds (CMBs) and localized non-hemorrhage iron deposits in the basal ganglia have been associated with brain aging, vascular disease and neurodegenerative disorders. Particularly, CMBs are small lesions and require multiple neuroimaging modalities for accurate detection. Quantitative susceptibility mapping (QSM) derived from in...
Contexts in source publication
Context 1
... annotation was performed according to a protocol developed with the focus on highly specific differential detection of CMBs and non-hemorrhage iron deposits based on multiple modalities including QSM. The detailed protocol is described in Sect. 2 in the "Supplementary materials", and a flowchart of the manual annotation process is shown in Supplementary Fig. 4 in the "Supplementary materials". Panel A in Fig. 4 shows an example of a CMB in the thalamus and non-hemorrhage iron deposits in the interior section of the globus pallidus on SWI (for TE = 7.5 ms and 22.5 ms), QSM and T2w MRI, and Panel B shows the expert segmentation of the lesions based on the annotation protocol. ...
Context 2
... with the focus on highly specific differential detection of CMBs and non-hemorrhage iron deposits based on multiple modalities including QSM. The detailed protocol is described in Sect. 2 in the "Supplementary materials", and a flowchart of the manual annotation process is shown in Supplementary Fig. 4 in the "Supplementary materials". Panel A in Fig. 4 shows an example of a CMB in the thalamus and non-hemorrhage iron deposits in the interior section of the globus pallidus on SWI (for TE = 7.5 ms and 22.5 ms), QSM and T2w MRI, and Panel B shows the expert segmentation of the lesions based on the annotation protocol. Panel C shows an example of a larger CMB located in the occipital ...
Citations
... As a contribution to this growing research, Isensee et al. (2021) developed nnU-Net, a self-configuring method for medical image segmentation that adapts based on the provided dataset. According to Rashid et al. (2021) deep learning can automatically segment cerebral microbleeds from structural brain MRI scans. Furthermore, Kermi et al. (2022) developed a multi-view CNN combining the advantages of 2D and 3D networks for glioma segmentation. ...
Introduction
In neurological diagnostics, accurate detection and segmentation of brain lesions is crucial. Identifying these lesions is challenging due to its complex morphology, especially when using traditional methods. Conventional methods are either computationally demanding with a marginal impact/enhancement or sacrifice fine details for computational efficiency. Therefore, balancing performance and precision in compute-intensive medical imaging remains a hot research topic.
Methods
We introduce a novel encoder-decoder network architecture named the Adaptive Feature Medical Segmentation Network (AFMS-Net) with two encoder variants: the Single Adaptive Encoder Block (SAEB) and the Dual Adaptive Encoder Block (DAEB). A squeeze-and-excite mechanism is employed in SAEB to identify significant data while disregarding peripheral details. This approach is best suited for scenarios requiring quick and efficient segmentation, with an emphasis on identifying key lesion areas. In contrast, the DAEB utilizes an advanced channel spatial attention strategy for fine-grained delineation and multiple-class classifications. Additionally, both architectures incorporate a Segmentation Path (SegPath) module between the encoder and decoder, refining segmentation, enhancing feature extraction, and improving model performance and stability.
Results
AFMS-Net demonstrates exceptional performance across several notable datasets, including BRATs 2021, ATLAS 2021, and ISLES 2022. Its design aims to construct a lightweight architecture capable of handling complex segmentation challenges with high precision.
Discussion
The proposed AFMS-Net addresses the critical balance issue between performance and computational efficiency in the segmentation of brain lesions. By introducing two tailored encoder variants, the network adapts to varying requirements of speed and feature. This approach not only advances the state-of-the-art in lesion segmentation but also provides a scalable framework for future research in medical image processing.
... However, understanding the relationship between microRNAs and diseases is intricate and fraught with difficulties. According to previous research, miRNAs can function as oncogenes or tumor suppressors based on the cellular context [2]. Their dual function complicates our understanding of their role in disease mechanisms. ...
In the broad and complex field of biological data analysis, researchers frequently gather information from a single source or database. Despite being a widespread practice, this has disadvantages. Relying exclusively on a single source can limit our comprehension as it may omit various perspectives that could be obtained by combining multiple knowledge bases. Acknowledging this shortcoming, we report on miRGediNET, a novel approach combining information from three biological databases. Our investigation focuses on microRNAs (miRNAs), small non-coding RNA molecules that regulate gene expression post-transcriptionally. We delve deeply into the knowledge of these miRNA's interactions with genes and the possible effects these interactions may have on different diseases. The scientific community has long recognized a direct correlation between the progression of specific diseases and miRNAs, as well as the genes they target. By using miRGediNET, we go beyond simply acknowledging this relationship. Rather, we actively look for the critical genes that could act as links between the actions of miRNAs and the mechanisms underlying disease. Our methodology, which carefully identifies and investigates these important genes, is supported by a strategic framework that may open up new possibilities for comprehending diseases and creating treatments. We have developed a tool on the Knime platform as a concrete application of our research. This tool serves as both a validation of our study and an invitation to the larger community to interact with, investigate, and build upon our findings. miRGediNET is publicly accessible on GitHub at https://github.com/malikyousef/miRGediNET, providing a collaborative environment for additional research and innovation for enthusiasts and fellow researchers.
... QSM could be useful to accurately identify true CMBs since it allows to separate diamagnetic tissues (with negative susceptibility, appearing hypointense) from paramagnetic tissues (with positive susceptibility, appearing hyperintense). On QSM images, CMBs appear hyperintense while diamagnetic mimics (e.g., calcifications) will appear hypointense (Rashid et al., 2021). ...
... table, generally deep-learning-based methods performed better compared to conventional machine learning methods. Also, the methods using multiple modalities or using phase information in addition to SWI (Ghafaryasl et al., 2012;Liu et al., 2019b;Al-Masni et al., 2020;Rashid et al., 2021) showed better results. In fact, Al-Masni et al. (2020) showed that using phase in addition to SWI images improves the cluster-wise TPR by 5.6% (with only SWI: 91.6% and with SWI and phase: 97.2%) and Rashid et al. (2021) achieved the best CMB detection performance by using T2-weighted, SWI and QSM modalities. ...
... Also, the methods using multiple modalities or using phase information in addition to SWI (Ghafaryasl et al., 2012;Liu et al., 2019b;Al-Masni et al., 2020;Rashid et al., 2021) showed better results. In fact, Al-Masni et al. (2020) showed that using phase in addition to SWI images improves the cluster-wise TPR by 5.6% (with only SWI: 91.6% and with SWI and phase: 97.2%) and Rashid et al. (2021) achieved the best CMB detection performance by using T2-weighted, SWI and QSM modalities. However, our proposed method uses a single modality (SWI or T2*-GRE), along with the FRST images (obtained from the input modality itself) and gives comparable results to state-of-the-art methods such as Liu et al. (2019b) and Al-Masni et al. (2020), and with lower FPavg compared to Dou et al. (2016). ...
Introduction
Cerebral microbleeds (CMBs) are associated with white matter damage, and various neurodegenerative and cerebrovascular diseases. CMBs occur as small, circular hypointense lesions on T2*-weighted gradient recalled echo (GRE) and susceptibility-weighted imaging (SWI) images, and hyperintense on quantitative susceptibility mapping (QSM) images due to their paramagnetic nature. Accurate automated detection of CMBs would help to determine quantitative imaging biomarkers (e.g., CMB count) on large datasets. In this work, we propose a fully automated, deep learning-based, 3-step algorithm, using structural and anatomical properties of CMBs from any single input image modality (e.g., GRE/SWI/QSM) for their accurate detections.
Methods
In our method, the first step consists of an initial candidate detection step that detects CMBs with high sensitivity. In the second step, candidate discrimination step is performed using a knowledge distillation framework, with a multi-tasking teacher network that guides the student network to classify CMB and non-CMB instances in an offline manner. Finally, a morphological clean-up step further reduces false positives using anatomical constraints. We used four datasets consisting of different modalities specified above, acquired using various protocols and with a variety of pathological and demographic characteristics.
Results
On cross-validation within datasets, our method achieved a cluster-wise true positive rate (TPR) of over 90% with an average of <2 false positives per subject. The knowledge distillation framework improves the cluster-wise TPR of the student model by 15%. Our method is flexible in terms of the input modality and provides comparable cluster-wise TPR and better cluster-wise precision compared to existing state-of-the-art methods. When evaluating across different datasets, our method showed good generalizability with a cluster-wise TPR >80 % with different modalities. The python implementation of the proposed method is openly available.
... It was performed, for example, in Ghafaryasl et al. (2012), Liu et al. (2019) and Chen et al. (2019). A QSM image can be generated using Morphology Enabled Dipol Inversion (MEDI) , like in Rashid et al. (2021). ...
... Padding is an artificial size change by the addition of a black frame to obtain a desired image size without applying resize. It was utilized by Ferlin et al. (2021), Rashid et al. (2021) and Stanley and Franklin (2022a). ...
... , Mikolajczyk and Grochowski (2018) and Paszke et al. (2019). Augmentation was used in Afzal et al. (2022), Doke et al. (2020), Ferlin et al. (2021), Gunter et al. (2018), Li et al. (2021), Momeni et al. (2021), Myung et al. (2021),Rashid et al. (2021) andStandvoss et al. (2018). ...
... In a study of embedded networked medical systems, an algorithm for the automatic detection of CH using CT image data was suggested by Jiang et al. [6] and obtained an excellent detection accuracy. Rashid et al. [7] proposed a deep learningbased segmentation approach and utilized T2-weighted images, susceptibility-weighted imaging, and quantitative susceptibility mapping to segment these different lesion types. The sensitivity and accuracy of the proposed algorithm were experimentally demonstrated to be usable in large-scale studies. ...
Medical diseases seriously affect human life and health, and a segmentation model that can effectively support doctors in making the correct diagnoses of medical disease images is needed. Multi-threshold image segmentation is famous for its simplicity and ease of implementation, but the choice of its threshold combination affects its performance, and traditional optimization algorithms fall into local optimality with significant time consumption when solving such problems. Therefore, metaheuristic algorithms have been applied to this field, but many have drawbacks, such as slow convergence, easy prematureness, and unbalanced performance when performing threshold selection. For instance, the Hunger Games Search (HGS) algorithm proposed last year is unsatisfactory regarding convergence accuracy and speed. Hence, an improved HGS (SCHGS) is proposed by combining the slime mould position update mechanism and chaotic optimal solution variation. The slime position update mechanism has a powerful exploration capability, which can help HGS increase the exploration of the search space and find the optimal solution as much as possible. On the other hand, the chaotic optimal solution variation strengthens the algorithm's local exploitation ability, which can effectively avoid falling into the local optimum. The experimental results on benchmark test functions indicate that the convergence performance of SCHGS is improved by 54% compared with the original algorithm, and there is a more obvious advantage in the convergence speed. In the application of threshold selection in brain hemorrhage image segmentation, the performance of the suggested method also improves by 0.08%, 0.55%, and 0.29% according to different evaluation metrics (FSIM, PSNR, and SSIM), further demonstrating the effectiveness of SCHGS in solving image segmentation problems.
... Various techniques could be used to improve classification performance, such as data augmentation, which is used to enhance performance by enlarging the training set (Rashid et al., 2021). In this work, we did not employ data augmentation as we were aiming to capture biology-informed patterns, and the most standard form of data augmentation, the inclusion of translated and rotated copies of the training scans (Rashid et al., 2021), would have blurred the boundaries of the brain regions that the CNN heatmaps were aiming to capture. ...
... Various techniques could be used to improve classification performance, such as data augmentation, which is used to enhance performance by enlarging the training set (Rashid et al., 2021). In this work, we did not employ data augmentation as we were aiming to capture biology-informed patterns, and the most standard form of data augmentation, the inclusion of translated and rotated copies of the training scans (Rashid et al., 2021), would have blurred the boundaries of the brain regions that the CNN heatmaps were aiming to capture. In the future, we will check whether the use of more advanced data augmentation methods, such as the introduction of realistic noises that preserve the boundaries between grey matter and white matter in the MRI scans, could be used to carry out a data augmentation that retains the boundaries of the regions of interest. ...
Deep neural networks currently provide the most advanced and accurate machine learning models to distinguish between structural MRI scans of subjects with Alzheimer's disease and healthy controls. Unfortunately, the subtle brain alterations captured by these models are difficult to interpret because of the complexity of these multi-layer and non-linear models. Several heatmap methods have been proposed to address this issue and analyze the imaging patterns extracted from the deep neural networks, but no quantitative comparison between these methods has been carried out so far. In this work, we explore these questions by deriving heatmaps from Convolutional Neural Networks (CNN) trained using T1 MRI scans of the ADNI data set and by comparing these heatmaps with brain maps corresponding to Support Vector Machine (SVM) activation patterns. Three prominent heatmap methods are studied: Layer-wise Relevance Propagation (LRP), Integrated Gradients (IG), and Guided Grad-CAM (GGC). Contrary to prior studies where the quality of heatmaps was visually or qualitatively assessed, we obtained precise quantitative measures by computing overlap with a ground-truth map from a large meta-analysis that combined 77 voxel-based morphometry (VBM) studies independently from ADNI. Our results indicate that all three heatmap methods were able to capture brain regions covering the meta-analysis map and achieved better results than SVM activation patterns. Among them, IG produced the heatmaps with the best overlap with the independent meta-analysis.
... Studies on the tools that automatically detect microbleeds can be divided into three categories based on the source of the data used. The first category includes researches that use data from specific, existing studies: [ [32,33], Traumatic Brain Injury (TBI) [45,29,5,44,140], stroke [31,5,73,20], Intracerebral Haemorrhages (ICH) [34,20], gliomas [26,51,17], hemodialysis cases [5], Cerebral Amyloid Angiopathy (CAA) [34], atherosclerosis [6], or did not distinguish any particular disease besides the appearance of CMBs [1,30,42,43,3,80,79,24,18,19,76,22,13,72,20,126,138,137]. Datasets used in the first category of researches focused on AD [81,82,83,36,84,85], SMART [37], TBI [86], stroke [86,89], ICH [87,90,91], gliomas [70], hemodialysis cases [86], CAA [88], atherosclerosis [38], or CMBs [92]. ...
... Nevertheless, there are also other methods for bias field correction [93,16]. This operation was applied by [21,28,27,44,45,46,29,5,125,6,74,20]. Skull stripping also known as brain extraction is an operation of removing skull and background from the image, leaving only the brain. ...
... This enables bias reduction in the next stages of system creation. It was applied by [32,73,25,33,35,26,41,28,27,42,43,45,5,6,18,72]. ...
Cerebral microbleeds detection is an important and challenging task. With the gaining popularity of the MRI, the ability to detect cerebral microbleeds also raises. Unfortunately, for radiologists, it is a time-consuming and laborious procedure. For this reason, various solutions to automate this process have been proposed for several years, but none of them is currently used in medical practice. In this context, the need to systematize the existing knowledge and best practices has been recognized as a factor facilitating the imminent synthesis of a real CMBs detection system practically applicable in medicine. To the best of our knowledge, all available publications regarding automatic cerebral microbleeds detection have been gathered, described, and assessed in this paper in order to distinguish the current research state and provide a starting point for future studies.
... Microbleeds were initially identified by a deep learning-based segmentation method that used T2-weighted images and quantitative susceptibility mapping/susceptibility weighted imaging to segment the lesions and to differentiate microbleeds from iron deposits. 18 Identified lesions were then reviewed by a radiologist (J.B.W.) who made the final classification. The 99 th percentile of the total microbleed count distribution was 12 microbleeds. ...
... A total of 1535 MESA participants completed ≥24 hours of ambulatory ECG monitoring with a median (interquartile range) monitoring duration of 14.0 (13.5, 26.1) days. Among these participants, 1010 completed the MRI of the brain a median (interquartile range) of 17 (15)(16)(17)(18)(19) months later, and 967 were free of a prior history of stroke or transient ischemic attack and had MRI images of the brain that met quality control criteria ( Figure 1). Characteristics at Exam 6 of these 967 participants are described in Table 1. ...
Background
Atrial fibrillation (AF) is associated with increased stroke risk and accelerated cognitive decline, but the association of early manifestations of left atrial (LA) impairment with subclinical changes in brain structure is unclear. We investigated whether abnormal LA structure and function, greater supraventricular ectopy, and intermittent AF are associated with small vessel disease on magnetic resonance imaging of the brain.
Methods and Results
In the Multi‐Ethnic Study of Atherosclerosis, 967 participants completed 14‐day ambulatory electrocardiographic monitoring, speckle tracking echocardiography and, a median 17 months later, magnetic resonance imaging of the brain. We assessed associations of LA volume index and reservoir strain, supraventricular ectopy, and prevalent AF with brain magnetic resonance imaging measures of small vessel disease and atrophy. The mean age of participants was 72 years; 53% were women. In multivariable models, LA enlargement was associated with lower white matter fractional anisotropy and greater prevalence of microbleeds; reduced LA strain, indicating worse LA function, was associated with more microbleeds. More premature atrial contractions were associated with lower total gray matter volume. Compared with no AF, intermittent AF (prevalent AF with <100% AF during electrocardiographic monitoring) was associated with lower white matter fractional anisotropy (−0.25 SDs [95% CI, −0.44 to −0.07]) and greater prevalence of microbleeds (prevalence ratio: 1.42 [95% CI, 1.12–1.79]).
Conclusions
In individuals without a history of stroke or transient ischemic attack, alterations of LA structure and function, including enlargement, reduced strain, frequent premature atrial contractions, and intermittent AF, were associated with increased markers of small vessel disease. Detailed assessment of LA structure and function and extended ECG monitoring may enable early identification of individuals at greater risk of small vessel disease.
... We aim to evaluate the accuracy and reliability of ePVS segmentation in the presence or absence of T2w MRI, and when T2w is combined with other MRI sequences. We used a variation of our method, previously developed using MESA brain MRI data for fully automated detection of cerebral microbleeds and non-haemorrhage iron deposits in the basal ganglia (Rashid et al., 2021), and investigate the optimal strategy of combining information from susceptibility weighted imaging (SWI), fluid-attenuated inversion recovery (FLAIR), T1w and T2w MRI sequences. A set of ePVS segmentations by a human expert served as the gold standard for model training. ...
... Suppose : ℝ × → ℝ is a nonlinear function with a set of learnable parameters , where is the number of MRI Sequences used and is the size of the images, maps the images to voxel-wise labels indicating whether the voxel contains ePVS or not. In this study, is implemented as a multi-channel deep neural network (Rashid et al., 2021), which is a variation of the standard U-Net (Ronneberger, Fischer, & Brox, 2015) and has been demonstrated superior compared to conventional U-Net for small lesions (Rashid et al., 2021). A typical U-Net is made up of a downsampling or encoding path and a symmetric up-sampling or decoding path. ...
... Suppose : ℝ × → ℝ is a nonlinear function with a set of learnable parameters , where is the number of MRI Sequences used and is the size of the images, maps the images to voxel-wise labels indicating whether the voxel contains ePVS or not. In this study, is implemented as a multi-channel deep neural network (Rashid et al., 2021), which is a variation of the standard U-Net (Ronneberger, Fischer, & Brox, 2015) and has been demonstrated superior compared to conventional U-Net for small lesions (Rashid et al., 2021). A typical U-Net is made up of a downsampling or encoding path and a symmetric up-sampling or decoding path. ...
Deep learning has been demonstrated effective in many neuroimaging applications. However, in many scenarios the number of imaging sequences capturing information related to small vessel disease lesions is insufficient to support data-driven techniques. Additionally, cohort-based studies may not always have the optimal or essential imaging sequences for accurate lesion detection. Therefore, it is necessary to determine which of these imaging sequences are essential for accurate detection. In this study we aimed to find the optimal combination of magnetic resonance imaging (MRI) sequences for deep learning-based detection of enlarged perivascular spaces (ePVS). To this end, we implemented an effective light-weight U-Net adapted for ePVS detection and comprehensively investigated different combinations of information from susceptibility weighted imaging (SWI), fluid-attenuated inversion recovery (FLAIR), T1-weighted (T1w) and T2-weighted (T2w) MRI sequences. We conclude that T2w MRI is the most important for accurate ePVS detection, and the incorporation of SWI, FLAIR and T1w MRI in the deep neural network could make insignificant improvements in accuracy.
... Various techniques could be used to improve classification performance, such as data augmentation, which is used to enhance performance by enlarging the training set [74]. In this work, we did not employ data augmentation as we were aiming to capture biology-informed patterns, and the most standard form of data augmentation, the inclusion of translated and rotated copies of the training scans [74], would have blurred the boundaries of the brain regions that the CNN heatmaps were aiming to capture. ...
... Various techniques could be used to improve classification performance, such as data augmentation, which is used to enhance performance by enlarging the training set [74]. In this work, we did not employ data augmentation as we were aiming to capture biology-informed patterns, and the most standard form of data augmentation, the inclusion of translated and rotated copies of the training scans [74], would have blurred the boundaries of the brain regions that the CNN heatmaps were aiming to capture. In the future, we will check whether the use of more advanced data augmentation methods, such as the introduction of realistic noises that preserve the boundaries between grey matter and white matter in the MRI scans, could be used to carry out a data augmentation that retains the boundaries of the regions of interest. ...
Deep neural networks currently provide the most advanced and accurate machine learning models to distinguish between structural MRI scans of subjects with Alzheimer's disease and healthy controls. Unfortunately, the subtle brain alterations captured by these models are difficult to interpret because of the complexity of these multi-layer and non-linear models. Several heatmap methods have been proposed to address this issue and analyze the imaging patterns extracted from the deep neural networks, but no quantitative comparison between these methods has been carried out so far. In this work, we explore these questions by deriving heatmaps from Convolutional Neural Networks (CNN) trained using T1 MRI scans of the ADNI data set, and by comparing these heatmaps with brain maps corresponding to Support Vector Machines (SVM) coefficients. Three prominent heatmap methods are studied: Layer-wise Relevance Propagation (LRP), Integrated Gradients (IG), and Guided Grad-CAM (GGC). Contrary to prior studies where the quality of heatmaps was visually or qualitatively assessed, we obtained precise quantitative measures by computing overlap with a ground-truth map from a large meta-analysis that combined 77 voxel-based morphometry (VBM) studies independently from ADNI. Our results indicate that all three heatmap methods were able to capture brain regions covering the meta-analysis map and achieved better results than SVM coefficients. Among them, IG produced the heatmaps with the best overlap with the independent meta-analysis.
