Magnetic Resonance Imaging Manifestations of Cerebral Small Vessel Disease: Automated Quantification and Clinical Application
Cerebral small vessel disease (CSVD) is a disorder of cerebral microvessels that leads to abnormalities visible on brain imaging. The neuroimaging features of CSVD, as outlined by the Standards for Reporting Vascular Changes on Neuroimaging (STRIVE), include recent small subcortical infarcts (RSSI), lacunes, white matter hyperintensities (WMH), perivascular spaces (PVS), cerebral microbleeds (CMB), and brain atrophy. These features are associated with various clinical consequences, including cognitive impairment, stroke, and dementia. With the advancement of neuroimaging techniques, automated quantification methods for these features have become increasingly important in both research and clinical settings. This article provides a comprehensive overview of the progress in automated quantification of CSVD imaging features, their clinical relevance, and their potential applications in clinical trials.
Introduction
CSVD is a major contributor to vascular or mixed dementia and is responsible for at least 20% of all strokes worldwide. The neuroimaging features of CSVD, such as RSSI, WMH, lacunes, PVS, CMB, and brain atrophy, are visible on conventional structural magnetic resonance imaging (MRI). These features have both shared and distinct clinical consequences. Automated quantification methods have been developed to improve the efficiency and reproducibility of CSVD imaging analysis. This article reviews the recent progress in the automated quantification of CSVD imaging features, their clinical implications, and their potential use as endpoints in clinical trials.
Automated Quantification of MRI Features of CSVD
Recent Small Subcortical Infarcts (RSSI)
RSSI, also known as lacunar stroke, refers to recent infarction in the territory of one perforating artery, causing about 25% of ischemic strokes. RSSIs can be identified on MRI diffusion-weighted imaging (DWI) as hyperintense lesions up to 20 mm in diameter. Traditionally, the quantification of RSSIs has been based on visual inspection or manual delineation. However, automated methods, particularly those using deep learning, have shown promise. For example, Zhang et al. applied a deep convolutional neural network (CNN) to segment acute ischemic lesions, achieving a Dice similarity coefficient (DSC) of 79.13% and a lesion-wise precision of 92.67%. This method is applicable to the segmentation of RSSIs, although it does not exclusively target them.
White Matter Hyperintensities (WMH)
WMHs are hyperintense lesions visible on fluid-attenuated inversion recovery (FLAIR) and T2-weighted MRI. The Fazekas scale is the most widely used visual rating method for WMH, dividing them into periventricular WMH (PWMH) and deep WMH (DWMH) based on location and lesion size. Automated quantification methods for WMH can be classified into unsupervised and supervised methods. Unsupervised methods use intensity features for clustering, while supervised methods rely on manual delineations as ground truth for training. Deep learning methods, such as CNNs, have achieved a DSC of up to 0.80. The performance of these methods varies depending on lesion load, with higher DSCs observed in patients with larger WMH burdens.
Lacunes
Lacunes are round or ovoid subcortical fluid-filled cavities visible on MRI, with diameters ranging from 3 to 15 mm. They appear as hypointensities on T1-weighted and FLAIR images and hyperintensities on T2-weighted images. Automated methods for lacune detection are limited, but deep learning approaches have shown promise. For example, Ghafoorian et al. used a deep CNN with information from FLAIR and T1-weighted images, achieving a sensitivity of 97.4% with 0.13 false positives per slice.
Cerebral Microbleeds (CMB)
CMBs are small areas of signal void on T2-weighted gradient-recalled echo (GRE) or susceptibility-weighted imaging (SWI). SWI is more reliable for CMB detection than T2GRE. Automated detection methods have evolved from using morphological features to deep learning-based approaches. Dou et al. applied a 3D CNN for CMB detection, achieving a sensitivity of 93.16% and a precision of 44.31%. The use of 7T SWI or phase images can further improve detection accuracy, but these techniques are less accessible in clinical practice.
Perivascular Spaces (PVS)
PVS, also known as Virchow-Robin spaces, are fluid-filled spaces that follow the course of cerebral penetrating vessels. Enlarged PVS (EPVS) become more apparent with age and are associated with other CSVD features. Automated quantification of EPVS is challenging due to their small size and varying appearance. Some studies have used deep learning methods for EPVS segmentation, achieving a DSC of 0.73. Others have developed semi-automatic methods or used visual ratings as ground truth for training.
Brain Atrophy
Brain atrophy can be assessed in specific lobes, tissues, or regions. Automated segmentation methods for brain atrophy include tissue segmentation (e.g., white matter, gray matter, and cerebrospinal fluid) and anatomical structure segmentation (e.g., hippocampus, thalamus). Deep learning methods have achieved DSCs of 0.85 to 0.90 for various brain structures. These methods are increasingly used in research and clinical settings due to their accuracy and efficiency.
Clinical Relevance of MRI Features of CSVD
Recent Small Subcortical Infarcts (RSSI)
RSSIs are often symptomatic and may evolve into lacunes or remain as non-cavitated WMHs. They are associated with motor dysfunction and cognitive impairment, depending on the lesion location. The initial mortality and stroke recurrence rates are low, and most patients recover well in the first few weeks after onset.
White Matter Hyperintensities (WMH)
WMHs are associated with cognitive impairment, dementia, and stroke recurrence. They also contribute to brain atrophy patterns seen in Alzheimer’s disease. The location and shape of WMHs can influence their clinical impact, with certain strategic locations and irregular shapes being more strongly associated with cognitive deficits.
Cerebral Microbleeds (CMB)
CMBs are associated with cognitive decline and an increased risk of dementia. The number and location of CMBs, particularly in deep or mixed regions, are significant predictors of cognitive impairment. In stroke patients, a high number of CMBs is associated with an increased risk of recurrent intracranial hemorrhage.
Lacunes
Lacunes, whether symptomatic or silent, are associated with cognitive impairment and an increased risk of stroke and dementia. The location of lacunes, particularly in the thalamus and basal ganglia, is important for predicting post-stroke depression and cognitive decline.
Perivascular Spaces (PVS)
EPVS are associated with cognitive decline, stroke recurrence, and post-stroke depression. The location of EPVS, particularly in the centrum semiovale or basal ganglia, is important for differentiating between Alzheimer’s disease and subcortical vascular cognitive impairment.
Brain Atrophy
Brain atrophy is associated with cognitive impairment and dementia. Regional atrophy, such as hippocampal atrophy, is a significant biomarker for Alzheimer’s disease. Brain atrophy is also associated with other CSVD features, such as WMH, CMB, lacunes, and EPVS.
Possibility as Endpoints in Clinical Trials
CSVD imaging features have been used as endpoints in clinical trials to evaluate the progression of CSVD and the efficacy of interventions. WMH is the most widely used feature, followed by brain atrophy, CMB, and lacunes. EPVS has not yet been used in clinical trials due to inconsistent findings and the lack of robust quantification tools. RSSI is primarily used in trials targeting stroke prevention rather than CSVD progression.
The Impact of Advanced Neuroimaging Quantification on CSVD
Advanced neuroimaging techniques and automated quantification methods have improved our understanding of CSVD mechanisms and prognosis. For example, the topographical association between lacunes and WMHs suggests a common underlying pathophysiology. Automated quantification methods have also enabled the development of more complex metrics for CSVD features, such as shape descriptors for lacunes and multidimensional metrics for EPVS. These metrics improve the sensitivity of detecting associations between CSVD features and clinical outcomes.
Conclusions
The robustness and efficiency of automated quantification methods for CSVD imaging features have significantly improved, particularly with the use of deep learning techniques. These methods provide detailed volumetric and locational information, which is essential for understanding CSVD mechanisms and improving prognosis. However, there is still a gap between the development of these methods and their application in clinical practice. Well-validated and easy-to-use automated tools that support the quantification of multiple CSVD features are needed to facilitate their translation into clinical use.
doi.org/10.1097/CM9.0000000000001299
Was this helpful?
0 / 0