Development and validation of morphological segmentation of age-related cerebral white matter hyperintensities

doi:10.1016/j.neuroimage.2009.03.055

NeuroImage

Volume 47, Issue 1, 1 August 2009, Pages 199-203

https://doi.org/10.1016/j.neuroimage.2009.03.055 Get rights and content

Abstract

Accurate automated segmentation of age-related white matter hyperintensity (WMH) is desirable for topological studies and those involving large samples. We assessed the accuracy of a novel automated method for segmentation of WMH on magnetic resonance imaging (MRI) in a randomly selected population-based sample of older people aged > 60 years. The method combined morphological segmentation and statistical classifiers. Validation of this method was performed against expert manual segmentation in a sample of 30 scans, and against semi-automated segmentation in 202 scans. Its performance was also compared with those of other known methods derived from simple thresholding or Gaussian mixture modelling.

Automated morphological segmentation combined with an adaptive boosting statistical classifier showed substantial agreement with manual segmentation, with an intraclass correlation coefficient (ICC) of 0.90 (95% confidence interval [CI], 0.80–0.95) for WMH volume and median similarity index (SI) of 0.58 (interquartile range [IQR] 0.50–0.65). The method also showed similarly high levels of agreement with semi-automated segmentation, with ICC 0.92 (95% CI 0.89–0.93) and median SI 0.56 (IQR 0.49–0.66). Its best performance was observed for the highest tertile of WMH volume. Threshold-based and Gaussian mixture model-driven automated segmentation generally did not perform well in this study.

Introduction

White matter hyperintensities (WMH) are visible on T2-weighted magnetic resonance imaging (MRI) brain scans of older people. They are known to have clinically important effects on cognition (Prins et al., 2004) and motor function (Guttmann and Benson, 2000, Rosano and Brach, 2006). They may be potential targets for intervention in the prevention or treatment of important public health problems such as dementia and falls. A variety of approaches have been developed for quantification of WMH, including simple visual rating scales (Fazekas et al., 1987) and more sophisticated manual, semi-automated (DeCarli and Miller, 1999, Sachdev and Parslow, 2004) and fully automated (Anbeek and Vincken, 2004, Admiraal-Behloul and van den Heuvel, 2005, Wu and Rosano, 2006, Dyrby and Rostrup, 2008) segmentation techniques.

Automated segmentation, a procedure that requires no user interaction at any stage, is the ideal way to achieve efficiency, particularly in large samples. Several approaches have been used for this purpose, including various forms of voxel-based classification, thresholding and region growing (a fuller summary of previously published methods is provided in Supplementary material, under Section 1.1 titled ‘Summary of previous segmentation methods’). Such methods are usually based on assumptions that explicitly or implicitly reject lesions that may be of interest, such as smaller or peripheral lesions. The rejection of such lesions has the potential to introduce error in WMH volume estimation or influence the results of topological studies designed to study the effect of lesion location on outcome.

We set out to develop a novel automated method of segmentation of WMH with the goal of automatically segmenting such problem lesions without degrading overall segmentation performance. In this approach, we applied morphological filtering and segmentation methods in combination with statistical classifiers, a two-phase approach that is unique to WMH segmentation. Morphological methods are a powerful suite of tools that provide a rich set of operators for low-level visual processing which allow a variety of cues, such as voxel brightness, edge strength and image topology to be combined in segmentation. Their application may allow all potential WMH regions, including those containing small or peripheral lesions, to be segmented consistently; thereby maximizing the information extracted from scans. Following the development of this method, we then comprehensively tested its performance against expert manual segmentation and semi-automated segmentation in a large population-based sample of older people. In addition, we compared its performance with that of other known automated segmentation methods.

Section snippets

Study sample

The sample consisted of the first 247 participants in the population-based Tasmanian Study of Cognition and Gait (TASCOG) conducted in Tasmania, Australia. These participants were aged between 60 and 86 years and randomly recruited from the Tasmanian electoral roll, a comprehensive list of residents. They were excluded if they lived in nursing homes, could not have an MRI, or had MRI evidence of stroke or other brain lesions. The Southern Tasmanian Human Research Ethics Committee approved the

Sample characteristics

The study sample characteristics are illustrated in Table 1. The mean WMH volumes according to tertiles in ascending order were 5.8 ml (SD = 1.1 ml), 9.9 ml (SD = 1.6 ml) and 25 ml (SD = 13.6 ml).

Validation of morphological segmentation

The performance of automated morphological segmentation varied according to the type of statistical classifier and feature vector used. Its combination with the adaptive boosting classifiers achieved generally good agreement with manual segmentation (Table 2). In particular, substantial agreement was

Discussion

We have developed and tested a novel and advanced morphological technique for the automated segmentation of age-related WMH designed to better identify smaller and more peripheral lesions. When used in combination with a ‘gentle’ adaptive boosting statistical lesion classifier, this method showed substantial agreement against manual segmentation. There was excellent agreement for estimation of WMH volume suggesting that its use for this purpose will achieve a good balance between efficiency and

Conclusions

Automated morphological segmentation of WMH is an accurate and valid method compared with expert manual segmentation. It provides a means of improving detection of small and peripheral WMH while achieving a good balance between accuracy and efficiency. Further manual editing may be important for detailed topological studies of the effect of WMH on outcome.

References (19)

Admiraal-BehloulF. et al.
Fully automatic segmentation of white matter hyperintensities in MR images of the elderly
NeuroImage
(2005)
AnbeekP. et al.
Probabilistic segmentation of white matter lesions in MR imaging
NeuroImage
(2004)
AshburnerJ. et al.
Unified segmentation
NeuroImage
(2005)
DyrbyT.B. et al.
Segmentation of age-related white matter changes in a clinical multi-center study
NeuroImage
(2008)
WenW. et al.
The topography of white matter hyperintensities on brain MRI in healthy 60- to 64-year-old individuals
NeuroImage
(2004)
WuM. et al.
A fully automated method for quantifying and localizing white matter hyperintensities on MR images
Psychiatry Res.
(2006)
Beare, R. and G. Lehmann (2005). ‘Finding regional extrema — methods and performance.’ Insight Journal...
Beare, R. and G. Lehmann (2006). ‘The watershed transform in ITK — discussion and new developments.’ Insight Journal...
DeCarliC. et al.
Predictors of brain morphology for the men of the NHLBI twin study
Stroke
(1999)

There are more references available in the full text version of this article.

Cited by (41)

Accuracy of TrUE-Net in comparison to established white matter hyperintensity segmentation methods: An independent validation study
2024, NeuroImage
White matter hyperintensities (WMH) are nearly ubiquitous in the aging brain, and their topography and overall burden are associated with cognitive decline. Given their numerosity, accurate methods to automatically segment WMH are needed. Recent developments, including the availability of challenge data sets and improved deep learning algorithms, have led to a new promising deep-learning based automated segmentation model called TrUE-Net, which has yet to undergo rigorous independent validation. Here, we compare TrUE-Net to six established automated WMH segmentation tools, including a semi-manual method. We evaluated the techniques at both global and regional level to compare their ability to detect the established relationship between WMH burden and age. We found that TrUE-Net was highly reliable at identifying WMH regions with low false positive rates, when compared to semi-manual segmentation as the reference standard. TrUE-Net performed similarly or favorably when compared to the other automated techniques. Moreover, TrUE-Net was able to detect relationships between WMH and age to a similar degree as the reference standard semi-manual segmentation at both the global and regional level. These results support the use of TrUE-Net for identifying WMH at the global or regional level, including in large, combined datasets.
Associations Between the Dietary Inflammatory Index, Brain Volume, Small Vessel Disease, and Global Cognitive Function
2021, Journal of the Academy of Nutrition and Dietetics
Citation Excerpt :
Volumes of hippocampi were determined based on manual tracings performed by a single expert using an established method.31 Fully automated WMH segmentation on T2 fluid-attenuated inversion recovery sequences32 was performed using a validated method.33 A voxel counting algorithm calculated WMH volume.
An inflammatory diet is related to poorer cognition, but the underlying brain pathways are unknown.
The aim of this study was to examine associations between the Energy-Adjusted Dietary Inflammatory Index (E-DII) and brain volume, small vessel disease, and cognition in people with and without type 2 diabetes mellitus (T2DM).
This is a secondary cross-sectional analysis of data from the Cognition and Diabetes in Older Tasmanians study.
This study included 641 participants (n = 326 with T2DM) enrolled between 2005 and 2011 from Tasmania, Australia.
The E-DII was computed from the 80-item Dietary Questionnaire for Epidemiological Studies, version 2. Brain volumes (gray matter, white matter, and white matter hyperintensities), infarcts, and microbleeds were obtained from magnetic resonance imaging. Global cognition was derived from a comprehensive battery of neuropsychological tests.
Logistic and linear regressions were performed to examine associations between E-DII and brain measures and a global cognitive score, adjusting for demographics, energy, T2DM, mood, ambulatory activity, and cardiovascular risk factors. An E-DII × T2DM interaction term was tested in each model.
The mean (standard deviation) age of participants was 69.8 (7.4) years. There were no associations between the E-DII and any of the brain structural measures or global cognitive function in fully adjusted models. There was a modification effect for T2DM on the association between E-DII and gray matter volume (T2DM: β = 1.38, 95% CI –3.03 to 5.79; without T2DM: β = –4.34, 95% CI, –8.52 to –0.16), but not with any of the other outcome measures.
In this cross-sectional study, E-DII was not associated with brain structure or global cognition. In 1 of the 7 outcomes, a significant modification effect for T2DM was found for the associations between E-DII and gray matter. Future prospective studies are needed to clarify the associations between diet-related inflammation and brain health.
A survey on U-shaped networks in medical image segmentations
2020, Neurocomputing
The U-shaped network is one of the end-to-end convolutional neural networks (CNNs). In electron microscope segmentation of ISBI challenge 2012, the concise architecture and outstanding performance of the U-shaped network are impressive. Then, a variety of segmentation models based on this architecture have been proposed for medical image segmentations. We present a comprehensive literature review of U-shaped networks applied to medical image segmentation tasks, focusing on the architectures, extended mechanisms and application areas in these studies. The aim of this survey is twofold. First, we report the different extended U-shaped networks, discuss main state-of-the-art extended mechanisms, including residual mechanism, dense mechanism, dilated mechanism, attention mechanism, multi-module mechanism, and ensemble mechanism, analyze their pros and cons. Second, this survey provides the overview of studies in main application areas of U-shaped networks, including brain tumor, stroke, white matter hyperintensities (WMHs), eye, cardiac, liver, musculoskeletal, skin cancer, and neuronal pathology. Finally, we summarize the current U-shaped networks, point out the open challenges and directions for future research.
Adherence to the Australian Dietary Guidelines Is Not Associated with Brain Structure or Cognitive Function in Older Adults
2020, Journal of Nutrition
Cognitive dysfunction is common in older adults, particularly in those with type 2 diabetes (T2D). Higher adherence to the Dietary Guidelines for Americans is associated with better brain health. However, it is unclear if adherence to the Australian Dietary Guidelines (ADG) is associated with cognition or brain structure in older adults.
The aims of this study were to 1) examine the relation between adherence to the ADG, cognition, and brain MRI and 2) determine whether T2D modifies any associations.
The Cognition and Diabetes in Older Tasmanians Study is a cross-sectional study in 688 people (n = 343 with T2D) aged 55–90 y. A validated 80-item food-frequency questionnaire was used to assess dietary intake. Adherence to the 2013 ADG was estimated using the Dietary Guidelines Index (DGI). Cognitive function in multiple domains was assessed with a comprehensive battery of neuropsychological tests and brain structure with MRI. Multivariable linear models were used to assess the associations between DGI, cognitive z scores, and brain structure. Effect modification for T2D was examined with a DGI × T2D product term.
The mean age of the sample was 69.9 y (SD: 7.4 y), with 57.1% men. The mean DGI was 54.8 (SD: 10.7; range: 24.1–84.6). No associations were observed between the Australian DGI and cognition or brain MRI measures. T2D did not modify any associations (P > 0.05).
This is the first study to investigate associations between adherence to the ADG and brain health in the older adults with and without T2D. Future prospective studies are required to clarify if there are long-term associations.
Deep convolutional neural network for accurate segmentation and quantification of white matter hyperintensities
2020, Neurocomputing
White matter hyperintensities (WMHs) appear as regions of abnormally signal intensity on magnetic resonance imaging (MRI) images, that can be identified in MRI images of elderly people and ischemic stroke patients. However, manual segmentation and quantification of images with WMHs is laborious and time-consuming. Moreover, ischemic stroke lesion and WMHs appear as similar signals in MRI images, making it difficult to accurately segment the WMHs. Analysis of WMH-containing images is important for clinical diagnosis, and thus several segmentation methods have been proposed. However, these methods cannot accurately differentiate WMH and ischemic stroke lesions. We propose a deep convolutional neural network, M2DCNN, that can accurately segment WMHs and distinguish them from ischemic stroke lesions. M2DCNN consists of two subnets that rely on a set of novel multi-scale features and a novel architecture (inclusion of dense and dilated blocks). Our model is trained and evaluated on two public segmentation challenges with multi-modality MRI images. Empirical tests demonstrate that M2DCNN outperforms current segmentation methods. We empirically demonstrate that M2DCNN effectively separates WMHs from stroke lesions. Finally, ablation experiments reveal that both multi-scale features and architectural elements in our method contribute to the improved predictive performance.
Dietary Patterns Are Not Associated with Brain Atrophy or Cerebral Small Vessel Disease in Older Adults with and without Type 2 Diabetes
2019, Journal of Nutrition
Background Unhealthy dietary patterns (DPs) are associated with poorer cognition, but few studies have investigated the underlying brain structural mechanisms.
Objective We aimed to examine the relations between DPs, brain structure, and cognition in older people with and without type 2 diabetes.
Methods This cross-sectional study consisted of a sample of people with (n = 343) and without type 2 diabetes (n = 346) aged 55–90 y. The 80-item Cancer Council of Victoria FFQ was used to assess dietary intake. Two DPs (prudent and traditional) for people with type 2 diabetes and 3 DPs (prudent, traditional, and Western) for those without type 2 diabetes were derived using principal component analysis. Neuropsychological tests assessed 6 cognitive domains. Brain MRI was performed to obtain gray, white matter, and hippocampal volumes and markers of small vessel disease (microbleeds, infarcts, and white matter hyperintensities). Multivariable linear regression was used to assess the cross-sectional associations between DPs, brain MRI, and cognitive variables.
Results For those without type 2 diabetes, higher adherence to the Western DP was associated with lower gray matter volume (β = −3.03 95% CI: −5.67, −0.38; P = 0.03). The addition of a cardiovascular risk score, mood, and physical activity weakened associations such that they were no longer significant (β = −1.97 (95% CI: −4.68, 0.74) P = 0.15) for the Western DP. There were no significant associations for the other DPs in people with and without type 2 diabetes.
Conclusion In this cross-sectional study, DPs were not independently associated with brain structure in people with or without type 2 diabetes. Future prospective studies are needed to clarify the role of vascular risk factors on associations between DPs and brain health.

View all citing articles on Scopus

View full text

Development and validation of morphological segmentation of age-related cerebral white matter hyperintensities

Abstract

Introduction

Section snippets

Study sample

Sample characteristics

Validation of morphological segmentation

Discussion

Conclusions

NeuroImage

NeuroImage

NeuroImage

NeuroImage

NeuroImage

Psychiatry Res.

Predictors of brain morphology for the men of the NHLBI twin study

Stroke