Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2018 Oct 27;40(4):1174–1183. doi: 10.1002/hbm.24437

Corticospinal tract degeneration in ALS unmasked in T1‐weighted images using texture analysis

Abdullah Ishaque 1,2, Dennell Mah 3, Peter Seres 4, Collin Luk 1, Wendy Johnston 3, Sneha Chenji 2, Christian Beaulieu 4, Yee‐Hong Yang 5, Sanjay Kalra 2,3,4,5,✉
PMCID: PMC6865626  PMID: 30367724

Abstract

The purpose of this study was to investigate whether textures computed from T1‐weighted (T1W) images of the corticospinal tract (CST) in amyotrophic lateral sclerosis (ALS) are associated with degenerative changes evaluated by diffusion tensor imaging (DTI). Nineteen patients with ALS and 14 controls were prospectively recruited and underwent T1W and diffusion‐weighted magnetic resonance imaging. Three‐dimensional texture maps were computed from T1W images and correlated with the DTI metrics within the CST. Significantly correlated textures were selected and compared within the CST for group differences between patients and controls using voxel‐wise analysis. Textures were correlated with the patients' clinical upper motor neuron (UMN) signs and their diagnostic accuracy was evaluated. Voxel‐wise analysis of textures and their diagnostic performance were then assessed in an independent cohort with 26 patients and 13 controls. Results showed that textures autocorrelation, energy, and inverse difference normalized significantly correlated with DTI metrics (p < .05) and these textures were selected for further analyses. The textures demonstrated significant voxel‐wise differences between patients and controls in the centrum semiovale and the posterior limb of the internal capsule bilaterally (p < .05). Autocorrelation and energy significantly correlated with UMN burden in patients (p < .05) and classified patients and controls with 97% accuracy (100% sensitivity, 92.9% specificity). In the independent cohort, the selected textures demonstrated similar regional differences between patients and controls and classified participants with 94.9% accuracy. These results provide evidence that T1‐based textures are associated with degenerative changes in the CST.

Keywords: adult, amyotrophic lateral sclerosis, corticospinal tracts, female, humans, magnetic resonance imaging, male, validation studies

1. INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder of the human motor system involving upper and lower motor neurons (UMN and LMN). There are no curative options and patients have a median survival of 26 months from the time of diagnosis (Pupillo, Messina, Logroscino, & Beghi, 2014). Clinical diagnosis of ALS is made in the presence of UMN and LMN physical exam findings and can be supported by electromyographic evaluation of subclinical LMN dysfunction (Brooks, Miller, Swash, & Munsat, 2000). Clinical magnetic resonance imaging (MRI) studies are used to rule out other diagnoses and do not provide an objective assessment of UMN dysfunction. On average, there is a delay of 9–16 months in reaching the diagnosis due to the inadequate sensitivity of clinical examination (Kraemer, Buerger, & Berlit, 2010).

Classic white matter pathology of ALS is characterized by the degeneration of the corticospinal tract (CST). Intensity variations within the CST have been demonstrated in conventional MRI sequences such as T1‐weighted (T1W) and fluid‐attenuated inversion recovery (FLAIR), but with subjective visual evaluation and low diagnostic performance (Bowen et al., 2000; Hecht et al., 2001). Recent attempts have also been made to improve the detection of pathological white matter intensity changes in T1W images using voxel‐based intensitometry (VBI) (Hartung et al., 2014). Regional cortical thinning on T1W images is a well‐characterized feature in ALS. Reductions in cortical thickness are observed in motor (precentral gyrus) and extra‐motor (frontotemporal) regions of the brain and are correlated with disease progression and functional disability (Agosta et al., 2012; Bede et al., 2013; Verstraete et al., 2012). The pathological correlates of these in vivo MRI findings, however, have not yet been established.

Diffusion tensor imaging (DTI) is an MRI modality that is used to study the microstructural changes in the central nervous system (Beaulieu, 2002). In vivo DTI studies have demonstrated altered water diffusion in the CST suggestive of microstructural pathology (Ciccarelli et al., 2006; Ciccarelli et al., 2009; Cosottini et al., 2005; Sarica et al., 2017). Advanced MRI modalities, such DTI, however, are not included routinely in clinical MRI studies and pose feasibility challenges including heterogeneity in sequences, long scan times, and availability on clinical scanners. Therefore, there is a need for an objective marker that is sensitive to the pathological UMN changes in ALS and that can be used clinically to mitigate delays in diagnosis and monitor disease progression.

Texture analysis is a statistical method of quantifying gray‐level intensity and patterns in an image that the human eye cannot detect. Studies have applied texture analysis in conventional MR images to characterize and diagnose various neurological diseases including brain tumors and Alzheimer's disease (Kickingereder et al., 2016; Sørensen et al., 2017). A novel three‐dimensional (3D) voxel‐wise texture analysis method also revealed the spatial pattern of pathology involving the medial temporal lobes in Alzheimer's disease in a hypothesis‐free study (Maani, Yang, & Kalra, 2015). In ALS, texture changes have been observed in the CST and the deep gray nuclei (Albuquerque et al., 2016; Maani, Yang, Emery, & Kalra, 2016). However, the pathological substrates that underlie texture changes have not yet been investigated.

The purpose of this study was to investigate whether textures of the CST quantified from T1W images are associated with in vivo surrogates of white matter degeneration in ALS. It was hypothesized that MRI textures of the CST from T1W images will correlate with DTI metrics of CST degeneration and the clinical UMN burden in patients. Furthermore, the diagnostic utility of T1‐based textures in ALS in two independent cohorts was also evaluated for replicability.

2. MATERIALS AND METHODS

Experiments were first conducted in patients with ALS and controls from a primary cohort to establish a set of textures from T1W images that correlate with DTI metrics. These textures were subsequently tested for voxel‐wise texture differences in the CST between patients and controls and evaluated for their diagnostic performance in the primary cohort and in an independent cohort for validation. A summary of the analytical steps of texture analysis in this study are presented in Supporting Information Figure E1.

2.1. Study participants

Patients with ALS were prospectively recruited from a multidisciplinary ALS clinic at the University of Alberta. Patients were required to have an El Escorial designation of clinically possible ALS and, as such, had UMN signs in at least one anatomical region (Brooks et al., 2000). Patients with co‐morbid frontotemporal dementia (FTD) and familial ALS were also eligible. Patients were excluded if they were older than 80 years of age, or if they had a history of co‐morbid neurological conditions including, but not limited to, multiple sclerosis, stroke, and brain trauma. Healthy control participants without neurological, or psychiatric disorders were recruited. Participants were excluded if they could not tolerate the duration of the MRI scan, or if they had contraindications to having an MRI such as a cardiac pacemaker. All participants provided informed written consent before participating in the study, which was approved by the local research ethics review board.

Nineteen patients and 14 controls were enrolled through the Canadian ALS Neuroimaging Consortium into the primary cohort. One patient presented with co‐morbid FTD and there were no cases of familial ALS. Patients underwent a neurological exam to assess the extent of clinical UMN involvement in the limbs and bulbar region. A UMN burden score with a maximum score of 12 was calculated. The presence of spasticity and hyperreflexia in the upper and lower limbs, including the Babinski sign and clonus in each foot, were assessed for a maximum possible score of 6 from each side of the body. Twenty‐six patients and 13 controls were enrolled into the independent cohort. Disability was quantified with the ALS functional rating scale‐revised (ALSFRS‐R). Neurological exam data and the ALSFRS‐R scores were collected on the same day as the MRI, or from the last clinical visit if it was within 4 weeks of the MRI scan. A summary of the participants' characteristics is provided in Table 1.

Table 1.

Participant characteristics of ALS patients and healthy controls

Primary cohort Replication cohort
Participant characteristic ALS patients Healthy controls ALS patients Healthy controls
Number of participants 19 14 26 13
Gender (n)
Male 11 6 18 5
Female 8 8 8 8
Age (years)
Mean ± SD 57.26 ± 10.11 56.93 ± 8.75 61.47 ± 9.22 53.62 ± 10.71
Range 37–74 37–67 42–77 28–65
Onset (n)
Bulbar 4 – 9 –
Limb 15 – 17 –
El Escorial clinical diagnosis category (n)
Definite ALS 3 – 9 –
Probable ALS 10 – 13 –
Probable ALS—Laboratory‐supported 0 – 4 –
Possible ALS 6 – 0 –
ALSFRS‐R (score)
Mean ± SD 41.00 ± 4.28 – 36.54 ± 4.93 –
Range 32–47 – 27–45 –
Symptom duration (months)
Mean ± SD 21.68 ± 14.26 – 19.58 ± 9.02 –
Range 9–60 – 2–38 –
Open in a new tab

Abbreviations: n = sample size; SD = standard deviation.

2.2. MRI parameters

Participants in the primary cohort were scanned on a Siemens Prisma 3T scanner. T1W images were acquired axially with a 3D magnetization‐prepared rapid gradient‐echo (MPRAGE) sequence (slice thickness = 1 mm, voxel size = 1 × 1 × 1 mm3, TR = 2,300 ms, TE = 3.4 s, flip angle = 9°, number of slices = 176). Diffusion‐weighted (DW) images were acquired axially using a two‐dimensional spin‐echo, single‐shot echo planar imaging (EPI) pulse sequence (slice thickness = 2 mm, voxel size = 2 × 2 × 2 mm3, TR = 10,000 ms, TE = 90 ms, flip angle = 90°, number of slices = 70, b0 images = 5, diffusion gradient directions = 30, b‐value = 1,000 s/mm2).

Participants in the independent cohort were scanned on a Siemens Sonata 1.5T scanner. T1W images were acquired axially with a 3D MPRAGE sequence (slice thickness = 1.5 mm, voxel size = 1 × 1 × 1.5 mm3, TR = 1,600 ms, TE = 3.8 s, flip angle = 15°, number of slices = 128).

2.3. T1W image analysis

T1W images were processed in statistical parametric mapping 8 (SPM8 revision 5236) (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/) running on MATLAB 2014b and were corrected for nonuniformity intensity bias. Gray and white matter were segmented and saved as a unified brain mask for each participant. T1W images were normalized to the Montreal Neurological Institute 152 (MNI152) template using diffeomorphic anatomical registration using exponentiated lie algebra (DARTEL) (Ashburner, 2007) and the forward and inverse deformation fields were saved for later transformations of texture maps into MNI space for voxel‐wise analysis. Gray and white matter images were smoothed with a 6 mm Gaussian kernel for voxel‐wise analysis.

Texture analysis was performed on the bias‐corrected T1W images in each participants' native space. 3D orthogonal texture maps were computed for each participant from their respective brain mask in SPM8 using a recently described method (Maani et al., 2015). The texture maps were computed using gray‐level co‐occurrence matrix (GLCM), a second‐order statistical method of examining textures (Haralick & Shanmugam, 1973). A GLCM was defined for each voxel and its adjacent voxels (referred to as the reference voxel and its neighborhood) in all three orthogonal planes for a T1W image. A GLCM is an N × N matrix, where N is the total number of gray levels in an image. To reduce computation time, T1W images were scaled down to 16 gray levels. Each cell in the GLCM (i, j) specifies the number of times gray level i co‐occurs with gray level j over a distance d and in a particular direction θ within the neighborhood. In this study, a distance of one pixel and four directions (0°, 45°, 90°, and 135°) adjacent to the central pixel were considered for the construction of the GLCMs. Eight directions (0°, ±45°, ±90°, ±135°, and 180°) can be considered in a GLCM; however, because of the symmetry found in the matrix across the diagonal, opposite directions (e.g., 0° and 180°) are redundant and were combined. The GLCMs for all four directions were summed and normalized to represent the probability of co‐occurrence between gray levels in the neighborhood. This was carried out in the axial, coronal, and sagittal planes to generate three GLCMs per voxel. A texture feature could then be calculated from the GLCMs and averaged over the three planes to compute a single 3D texture value for each voxel. Figure 1 illustrates a schematic representation of the creation of the GLCM. The following 18 texture feature maps were then computed from the GLCM: autocorrelation, contrast, correlation, cluster prominence, cluster shade, dissimilarity, energy, entropy, homogeneity, maximum probability, sum of squares, sum average, sum variance, sum entropy, difference variance, difference entropy, information measure of correlation, and inverse difference normalized. See Maani et al. (2015) for a full technical description and details regarding this method. Subsequent analyses of the CST were carried out within an atlas‐based CST mask as described in the later sections.

Figure 1.

Figure 1

Open in a new tab

Correlations were observed between DTI metrics of the CST and the texture features autocorrelation, energy, and inverse difference normalized in the primary cohort. Abbreviations: AD = axial diffusivity; FA = fractional anisotropy; FDR = false discovery rate correction; RD = radial diffusivity

2.4. DTI analysis

DW images were processed in ExploreDTI version 4.8.6 and were corrected for temporal signal drift, Gibbs ringing artifacts, motion, eddy current‐induced geometric distortions, and EPI distortions (Leemans, Jeurissen, Sijbers, & Jones, 2009; Perrone et al., 2015; Vos et al., 2017). EPI distortions were corrected through nonrigid registration of each participant's DWI data to their respective T1W image. Fractional anisotropy (FA), axial diffusivity (AD), and radial diffusivity (RD) maps were computed from the preprocessed DWI data.

Streamline tractography was performed with the following parameters: FA threshold = 0.2, fiber length range = 50–500 mm, and angle threshold = 30° using a deterministic approach (Basser, Pajevic, Pierpaoli, Duda, & Aldroubi, 2000). Regions of interests (ROIs) for the CST were developed on the MNI152 template. Bilateral “AND” operators in the cerebral peduncles and the precentral gyri were used to extract the CST. Contaminating corticopontine tracts and commissural tracts were excluded by placing “NOT” operators in the cerebellum, medial lemniscus, and corpus callosum that were obtained from the MNI structural atlas (Collins, Holmes, Peters, & Evans, 1995; Mazziotta et al., 2001) and the International Consortium of Brain Mapping DTI‐81 atlas (Hua et al., 2008). The ROIs were reversed transformed to each participant's native space using their respective inverse deformation fields. The resultant CSTs were segmented between the “AND” ROIs in each participant for anatomical consistency.

Segmented CSTs were resampled to 45 equidistant points for along‐tract analysis (Colby et al., 2012). Each unilateral CST bundle was averaged in cross section along its trajectory at each point. FA, AD, and RD values were computed from the 45 points along the single tract trajectories.

2.5. Statistical analysis

Statistical analyses were performed using IBM SPSS for Windows, Version 24.0. Highly correlated textures within the CST were removed to reduce computational load and redundancy. Specifically, a bilateral 3D CST mask was created for each participant by reversed transforming the CST mask from the Jülich histological atlas (Eickhoff et al., 2005) at 25% threshold. Texture values were averaged from the mask and Pearson's r was calculated between every pair of textures. One texture was removed from each pair if Pearson's r ≥ .90.

Textures were validated for sensitivity to white matter properties of the CST by correlating them with DTI metrics of the CST. Each participant's CST mask from the atlas was overlaid on their FA, AD, and RD maps to obtain averaged DTI metrics. Pearson's r was calculated between each texture and DTI metric pair and textures were selected for further analysis if a significant correlation was observed. Statistical significance was accepted at p < .05 corrected for multiple comparisons using the false discovery rate method (FDR), or at uncorrected p < .05 if the results did not reach significance after multiple comparison correction. The selected texture feature maps were transformed to the MNI space by applying the participants' respective forward deformation field and smoothed with a 6 mm Gaussian kernel for voxel‐wise analyses. Texture maps are inherently slightly blurry when calculated. Therefore, to avoid washing out texture details, a Gaussian kernel less than 8 mm, which is typically used in voxel‐wise studies, was used here.

Voxel‐wise two‐sample t‐tests were conducted for the microstructural‐sensitive textures between the two groups. Age was included as a covariate in all analyses. CST ROI from the atlas was used to create an explicit a priori mask in the MNI space. Significance was set at FDR p < .05, or uncorrected p < .01. A cluster size threshold of five or more voxels was set to report significant regions. Texture values from the significant clusters in each hemisphere were averaged for all patients and correlated with their contralateral UMN score using Spearman's rank correlation with a statistical significance of FDR p < .05. The reported textures from the CST were entered in a binary logistic regression model. The model was evaluated for diagnostic performance using receiver operating characteristic (ROC) curve analysis. Whole‐brain voxel‐based morphometry (VBM) analyses with two‐sample t‐tests to evaluate morphologic changes in gray and white matter were also conducted. Age was included as a covariate in both analyses. Statistical significance was accepted at FDR p < .05, or uncorrected p < .001 with a cluster size threshold of 20 or more.

Along‐tract analysis of DTI metrics was performed on unilateral CSTs. Between‐group differences in FA, AD, and RD were investigated individually by performing multivariate analyses of covariance with the DTI metric value at each location (45 values along the tract) as a dependent variable, group assignment (patients vs. controls) as an independent variable, and age as a covariate. To identify the location along the CST responsible for significant between‐group differences, comparisons were conducted using two‐tailed Student's t‐tests at each location adjusted for age with a statistical significance of FDR p < .05. Voxel‐wise analysis and the predictive model from the primary cohort were applied to the independent cohort to assess for reproducibility of results.

3. RESULTS

3.1. Correlation of DTI metrics with MRI textures

Ten textures remained after removing redundancies. Of the 10 textures, three demonstrated significant correlations with DTI metrics in the CST (Figure 1): autocorrelation correlated with AD (r = −.613) (FDR p < .05) and RD (r = −.556) (FDR p < .05); energy correlated with FA (r = .416) (uncorrected p < .05); and inverse difference normalized correlated with FA (r = .510) (FDR p < .05), AD (r = −.492) (FDR p < .05), and RD (r = −.675) (FDR p < .05).

3.2. Between‐group texture differences in CST

Voxel‐wise analysis revealed significant between‐group texture changes in the CST. Specifically, autocorrelation was increased in patients (uncorrected p < .01), whereas energy and inverse difference normalized were decreased in patients (FDR p < .05) (Figure 2). Left posterior limb of the internal capsule and bilateral centrum semiovale demonstrated differences in all three textures. Energy and inverse difference normalized demonstrated greater involvement of the left CST and bilateral posterior limbs of the internal capsule. Detailed regional analysis is reported in Supporting Information Table E1.

Figure 2.

Figure 2

Open in a new tab

Differences were present in textures in T1W images within the CST between patients and controls in the primary cohort overlaid on a sample T1W image. Abbreviations: FDR = false discovery rate correction, L = left, R = right [Color figure can be viewed at http://wileyonlinelibrary.com]

3.3. Association of textures in CST with clinical UMN burden

Significant correlations were found between textures in the CST from each hemisphere and the contralateral UMN burden score. Autocorrelation (r = .407) (FDR p < .05) and energy (r = −.382) (FDR p < .05) correlated with the contralateral UMN burden score, whereas inverse difference normalized did not demonstrate a significant correlation (Figure 3).

Figure 3.

Figure 3

Open in a new tab

Correlations were observed between textures autocorrelation and energy and the contralateral UMN score among patients in the primary cohort. Abbreviations: FDR = false discovery rate correction; UMN = upper motor neuron

3.4. Diagnostic performance of texture analysis

Textures autocorrelation, energy, and inverse difference normalized were entered in a regression model (predicted value = autocorrelation*.81 + energy*(−.22) + inverse difference normalized*(−3.36) + 599.47) and demonstrated 100% sensitivity, 92.9% specificity, and an overall classification accuracy of 97.0%. ROC curve analysis revealed an area under the ROC curve (AUC) of .985 (Figure 4).

Figure 4.

Figure 4

Open in a new tab

Receiver operating characteristic (ROC) curves for the regression models incorporating autocorrelation, energy, and inverse difference normalized in the primary cohort (left) and the independent cohort (right). The area under the curve (AUC) was .985 in the primary cohort (100% sensitivity and 92.9% specificity) and .991 in the independent cohort (92.3% sensitivity and 100% specificity) using the predictive model generated from the primary cohort

3.5. Between‐group VBM analyses

Gray matter VBM analysis revealed significant changes in the motor cortex bilaterally (uncorrected p < .001). Additional regional differences were observed in the temporal regions bilaterally and the left insula (Supporting Information Figure E2). White matter VBM analysis revealed significant subcortical changes in the right postcentral and precentral region, left parietal regions (uncorrected p < .001) (Supporting Information Figure E1). No significant differences were observed along the CST. Detailed regional analysis is reported in Supporting Information Table E2.

3.6. Along‐tract diffusion analysis

One patient's left CST was excluded from along‐tract analysis because tractography did not yield any tracts. Significant between‐group differences were demonstrated in bilateral CST FA (left: F = 14.0, df = 1, p < .05; right: F = 12.6, df = 1, p < .05) and RD (left: F = 17.8, df = 1, p < .05; right: F = 14.4, df = 1, p < .05). No significant between‐group difference was detected in AD. Along‐tract analysis revealed FA was significantly decreased in patients in the CST bilaterally in the posterior limb of the internal capsule and the centrum semiovale. RD was significantly increased in patients along most of the length of the left CST, and in the posterior limb of the internal capsule and centrum semiovale regions of the right CST (Figure 5).

Figure 5.

Figure 5

Open in a new tab

Along‐tract analysis of FA (top) and RD (bottom) in the left and the right CST for participants in the primary cohort. A sample CST constructed from tractography is superimposed on a sample T1W image to provide anatomical reference to the CST profiles. The x axes represent the diffusion metric and the y axes represent the position along the tract. Filled circles indicate points along the tract where the diffusion metric is significantly different in patients compared to controls. Reductions in FA in patients are localized mostly to the internal capsule, whereas RD is increased along most of the left CST and internal capsule and centrum semiovale of the right CST. Abbreviations: CST = corticospinal tract; FA = fractional anisotropy; FDR = false discover rate correction; RD = radial diffusivity [Color figure can be viewed at http://wileyonlinelibrary.com]

3.7. Independent cohort

Voxel‐wise analyses in the independent cohort revealed significant between‐group differences of the T1‐based textures similar to the primary cohort: autocorrelation was increased in patients (FDR p < .05) and energy and inverse difference normalized were decreased in patients (FDR p < .05) (Figure 6). Common areas of change in all three textures were the bilateral posterior limbs of the internal capsule and bilateral centrum semiovale. Energy and inverse difference normalized were also different in the cerebral peduncles bilaterally. Detailed regional analysis is reported in Supporting Information Table E3. Figure 7 shows a representative autocorrelation map and the corresponding T1‐weighted image from a control and a patient from this cohort. Application of the regression model derived from the primary cohort with the same weights achieved an overall classification accuracy of 94.9% with 92.3% sensitivity, 100% specificity, and an AUC of .991 (Figure 4).

Figure 6.

Figure 6

Open in a new tab

Differences in textures from T1W images within the CST between patients and controls in the independent cohort overlaid on a sample T1W image. Abbreviations: FDR = false discovery rate correction, L = left, R = right [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 7.

Figure 7

Open in a new tab

Representative 3D texture map for autocorrelation (top row) and the corresponding T1W image (bottom row) from a control (left) and a patient (right) in the independent cohort. In the control participant, the CST is homogenously hypointense on an autocorrelation map. In a patient, however, the CST appears to be more heterogenous with hyperintense areas in the centrum semiovale and in posterior limb of the internal capsule. The T1W image, in comparison to the texture map, shows no overt differences between the control and patient [Color figure can be viewed at http://wileyonlinelibrary.com]

4. DISCUSSION

In this study, it was shown that texture features of the CST in T1W images correlate with DTI indices, have discriminatory power in identifying ALS patients from controls, and correlate with clinical measures of UMN dysfunction. This suggests that degeneration of the CST in ALS can be evaluated using texture analysis of T1W images.

Textures from T1W images correctly classified patients and controls with 97.0% (100% sensitivity, 92.9% specificity) and 94.9% (92.3% sensitivity, 100% specificity) accuracy in two independent cohorts. This strong discriminatory ability of texture analysis in ALS was observed in a previous study that obtained 95% sensitivity and 90% specificity using textures from the CST in T1W images (Maani et al., 2016). Significant correlations were found between the textures and the UMN burden score indicating that changes detected in vivo by texture analysis of T1W images are associated with the underlying pathology of ALS.

T1‐based textures of the CST significantly correlated with DTI metrics. Animal models show that FA, AD, and RD are mediated by microstructural factors and processes, such as axonal membranes, axonal damage, and myelin damage, respectively (Beaulieu, 2002; Song et al., 2005). This suggests that texture changes from T1W images are sensitive to the characteristic white matter damage of the CST in ALS. The correlation between T1‐based textures and DTI metrics has been previously reported, although its biological implications have not been explored (Holli et al., 2010). Current analytical methods for T1W images, including VBM, are not sensitive to subtle intensity changes in white matter that may be mediated by altered microstructural properties. This was corroborated by our results (Supporting Information Figure E2). VBI is a contrast‐enhancement technique designed to overcome the current limitations of T1W VBM in evaluating white matter changes. A recent study using VBI in ALS demonstrated widespread white matter changes on T1W images, including along the CST (Hartung et al., 2014). This shows that pathological variations do exist on T1W images that are not detected with current VBM methods. Therefore, texture analysis can potentially enable white matter analysis from T1W images that reflect pathological alterations.

Textures also demonstrated differences between patients and controls in the centrum semiovale and the posterior limb of the internal capsule. Along‐tract analysis of DTI metrics also showed similar regions that were affected in patients compared to controls. Furthermore, FA and RD were altered in patients, whereas AD showed no change. This pattern of change in DTI metrics of the CST has been previously reported in ALS (Blain et al., 2011; Cosottini et al., 2005). Evidence suggests that it is characteristic of Wallerian axonal degeneration involving gliosis and increases in extracellular matrix (Beaulieu, 2002; Pierpaoli et al., 2001). Further evidence from combined MRI‐histopathology experiments suggests that T1 signal alterations in ALS are contributed, at least in part, by neuronal loss, myelin dysfunction, and astrocytosis (Meadowcroft et al., 2015). Studies in multiple sclerosis have demonstrated that texture analysis of T2W images is sensitive to the in vivo pathological acute inflammatory processes that occur in white matter lesions (Zhang, Zhu, Mitchell, Costello, & Metz, 2009). Zhang et al. (2009) reported that local coarse texture measured using the polar Stockwell transform technique from T2W images increases in acute inflammatory lesions during gadolinium enhancement. They also showed that pre‐lesion normal appearing white matter (NAWM) had similar texture to NAWM (Zhang et al., 2009). Ex vivo MRI‐pathology studies have further demonstrated that textures from MRI are directly related to tissue pathology. Higher texture heterogeneity was noted in T2W ex vivo MRI in regions of increasing burden of pathology in multiple sclerosis characterized by NAWM, diffusively abnormal white matter, and lesion (Zhang et al., 2013). Furthermore, Zhang et al. (2013) showed that the degree of texture heterogeneity in T2W images predicted the extent of pathological demyelination in histological analysis. Taken together, the findings of this study and the literature indicate that texture analysis in T1W images is likely detecting a substrate of pathological changes of the CST in ALS; however, MRI‐histopathology validation studies are needed in to verify this claim. This is important because T1W images are routinely acquired in the diagnostic workup of suspected patients to rule out gross structural abnormalities without any role in detecting disease pathology. With texture analysis, clinically acquired T1W images may be able to provide objective evidence of UMN involvement during the diagnostic workup.

Recent studies have proposed multimodal MRI as a means of increasing diagnostic yield over the use of a single imaging modality. A study combining DTI and MR spectroscopy reported 93% sensitivity and 85% specificity in discriminating patients from controls (Foerster et al., 2014). In another multimodal study, an 86% sensitivity and 78% accuracy was achieved in discriminating patients and controls using DTI metrics and gray matter densities from T1W images (Schuster, Hardiman, & Bede, 2016). Therefore, although studies support the diagnostic utility of multimodal MRI in ALS, texture analysis on T1W images alone has high diagnostic performance and with greater clinical feasibility as it does not require advanced MRI sequences with long acquisition times that may not be available on all scanners.

T1W images from the primary cohort were acquired from a 3T scanner, whereas the independent cohort was scanned in a 1.5T scanner. This has two important implications. First, texture analysis can be applied clinically without requiring high‐field MRI scanners. Second, the use of a common predictive model between the two cohorts highlights the robustness of texture analysis considering different MRI scanner vendors and acquisition parameters were used, an issue commonly present in multicentre studies and clinical trials. In addition, although ALS is pathologically characterized by UMN and LMN degeneration, it has substantial phenotypic heterogeneity and potential biomarkers must demonstrate diagnostic value across disease phenotypes. In this study, texture analysis demonstrated high diagnostic performance in distinguishing groups of heterogeneous patients from controls in two independent cohorts. Altogether, these results signify the clinical applicability and the utility of the use of textures analysis in ALS.

The study has several limitations. First, it consisted of two relatively small cohorts. Although the diagnostic performance of textures was replicated in an independent cohort, the observations made in this study need to be replicated in studies with larger sample sizes including age‐ and gender‐matched controls. Second, patients in this study comprised only those with an established diagnosis of ALS. Further studies should investigate the performance of texture analysis in discriminating ALS‐mimics and in identifying early‐onset patients in whom a diagnosis is uncertain.

In conclusion, this study demonstrates the utility of texture analysis in T1W images in detecting degeneration of the CST in ALS. It also provides clinical validity of texture‐based assessment of the CST in ALS given the clinical correlations and high discriminatory performance in independent cohorts. This study supports further analysis of textures from T1W images as MRI‐based biomarkers for UMN involvement in ALS.

CONFLICT OF INTEREST

The authors declare no conflict of interests.

Supporting information

Figure E1 Flow chart depicting a summary of the analytical steps involved in texture analysis. Steps: 1) 22 textures were calculated, 2) redundant textures were removed, 3) white matter‐sensitive textures were identified by correlating the unique textures to DTI metrics, 4) white matter‐sensitive textures were used for voxel‐wise analysis in both the primary and the independent cohorts, 5) texture values from the primary cohort were used to generate a binary regression model, and 6) used to evaluate the diagnostic performance of the white matter‐sensitive textures in both cohorts.

Figure E2: Gray (top) and white (bottom) matter VBM analyses for participants in the primary cohort. Significant differences at p < .001 with a cluster size threshold of 20 or more voxels were found. The results did not survive multiple comparison correction with FDR. VBM results are overlaid on a sample T1W image in neurological orientation. Detailed results are provided in Table E2. FDR = false discovery rate

Table E1: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure 2.

Table E2: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure E1.

Table E3: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure 6.

Click here for additional data file. (287.8KB, docx)

ACKNOWLEDGMENTS

This study was supported by funding from ALS Association, ALS Society of Canada, and Canadian Institutes of Health Research, and Brain Canada. Data were made available in part from the Canadian ALS Neuroimaging Consortium (CALSNIC), for which data management and quality control were facilitated by the Canadian Neuromuscular Disease Registry (CNDR).

Ishaque A, Mah D, Seres P, et al. Corticospinal tract degeneration in ALS unmasked in T1‐weighted images using texture analysis. Hum Brain Mapp. 2019;40:1174–1183. 10.1002/hbm.24437

Funding information: Brain Canada; Canadian Institutes of Health Research; ALS Society of Canada; ALS Association

REFERENCES

  1. Agosta, F. , Valsasina, P. , Riva, N. , Copetti, M. , Messina, M. J. , Prelle, A. , … Filippi, M. (2012). The cortical signature of amyotrophic lateral sclerosis. PLoS One, 7(8), e42816. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Albuquerque, M. , Anjos, L. G. , Andrade Helen, M. T. d. , Oliveira, M. S. , Castellano, G. , Junqueira Ribeiro de Rezende, T. , … Cavalcante França Junior, M. (2016). MRI texture analysis reveals deep gray nuclei damage in amyotrophic lateral sclerosis. Journal of Neuroimaging, 26(2), 201–206. [DOI] [PubMed] [Google Scholar]
  3. Ashburner, J. (2007). A fast diffeomorphic image registration algorithm. NeuroImage, 38(1), 95–113. [DOI] [PubMed] [Google Scholar]
  4. Basser, P. J. , Pajevic, S. , Pierpaoli, C. , Duda, J. , & Aldroubi, A. (2000). In vivo fiber tractography using DT‐MRI data. Magnetic Resonance in Medicine, 44(4), 625–632. [DOI] [PubMed] [Google Scholar]
  5. Beaulieu, C. (2002). The basis of anisotropic water diffusion in the nervous system–a technical review. NMR in Biomedicine, 15(7–8), 435–455. [DOI] [PubMed] [Google Scholar]
  6. Bede, P. , Bokde, A. , Elamin, M. , Byrne, S. , McLaughlin, R. L. , Jordan, N. , … Hardiman, O. (2013). Grey matter correlates of clinical variables in amyotrophic lateral sclerosis (ALS): A neuroimaging study of ALS motor phenotype heterogeneity and cortical focality. Journal of Neurology, Neurosurgery, and Psychiatry, 84(7), 766–773. 10.1136/jnnp-2012-302674 [DOI] [PubMed] [Google Scholar]
  7. Blain, C. R. , Brunton, S. , Williams, V. C. , Leemans, A. , Turner, M. R. , Andersen, P. M. , … Simmons, A. (2011). Differential corticospinal tract degeneration in homozygous 'D90A' SOD‐1 ALS and sporadic ALS. Journal of Neurology, Neurosurgery, and Psychiatry, 82(8), 843–849. 10.1136/jnnp.2010.236018 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bowen, B. C. , Pattany, P. M. , Bradley, W. G. , Murdoch, J. B. , Rotta, F. , Younis, A. A. , … Quencer, R. M. (2000). MR imaging and localized proton spectroscopy of the precentral gyrus in amyotrophic lateral sclerosis. AJNR–American Journal of Neuroradiology, 21(4), 647–658. [PMC free article] [PubMed] [Google Scholar]
  9. Brooks, B. R. , Miller, R. G. , Swash, M. , & Munsat, T. L. (2000). El Escorial revisited: Revised criteria for the diagnosis of amyotrophic lateral sclerosis. Amyotrophic Lateral Sclerosis and Other Motor Neuron Disorders, 1(5), 293–299. [DOI] [PubMed] [Google Scholar]
  10. Ciccarelli, O. , Behrens, T. , Altmann, D. , Orrell, R. , Howard, R. , Johansen‐Berg, H. , … Thompson, A. (2006). Probabilistic diffusion tractography: A potential tool to assess the rate of disease progression in amyotrophic lateral sclerosis. Brain, 129(7), 1859–1871. [DOI] [PubMed] [Google Scholar]
  11. Ciccarelli, O. , Behrens, T. E. , Johansen‐Berg, H. , Talbot, K. , Orrell, R. W. , Howard, R. S. , … Thompson, A. J. (2009). Investigation of white matter pathology in ALS and PLS using tract‐based spatial statistics. Human Brain Mapping, 30(2), 615–624. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Colby, J. B. , Soderberg, L. , Lebel, C. , Dinov, I. D. , Thompson, P. M. , & Sowell, E. R. (2012). Along‐tract statistics allow for enhanced tractography analysis. NeuroImage, 59(4), 3227–3242. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Collins, D. L. , Holmes, C. J. , Peters, T. M. , & Evans, A. C. (1995). Automatic 3‐D model‐based neuroanatomical segmentation. Human Brain Mapping, 3(3), 190–208. [Google Scholar]
  14. Cosottini, M. , Giannelli, M. , Siciliano, G. , Lazzarotti, G. , Michelassi, M. C. , Del Corona, A. , … Murri, L. (2005). Diffusion‐tensor MR imaging of corticospinal tract in amyotrophic lateral sclerosis and progressive muscular atrophy. Radiology, 237(1), 258–264. [DOI] [PubMed] [Google Scholar]
  15. Eickhoff, S. B. , Stephan, K. E. , Mohlberg, H. , Grefkes, C. , Fink, G. R. , Amunts, K. , & Zilles, K. (2005). A new SPM toolbox for combining probabilistic cytoarchitectonic maps and functional imaging data. NeuroImage, 25(4), 1325–1335. [DOI] [PubMed] [Google Scholar]
  16. Foerster, B. R. , Carlos, R. C. , Dwamena, B. A. , Callaghan, B. C. , Petrou, M. , Edden, R. A. , … Feldman, E. L. (2014). Multimodal MRI as a diagnostic biomarker for amyotrophic lateral sclerosis. Annals of Clinical and Translational Neurology, 1(2), 107–114. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Haralick, R. M. , & Shanmugam, K. (1973). Textural features for image classification. IEEE Transactions on Systems, Man, and Cybernetics, SMC‐3(6), 610–621. [Google Scholar]
  18. Hartung, V. , Prell, T. , Gaser, C. , Turner, M. R. , Tietz, F. , Ilse, B. , … Grosskreutz, J. (2014). Voxel‐based MRI intensitometry reveals extent of cerebral white matter pathology in amyotrophic lateral sclerosis. PLoS One, 9(8), e104894. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Hecht, M. , Fellner, F. , Fellner, C. , Hilz, M. , Heuss, D. , & Neundörfer, B. (2001). MRI‐FLAIR images of the head show corticospinal tract alterations in ALS patients more frequently than T2‐, T1‐and proton‐density‐weighted images. Journal of the Neurological Sciences, 186(1), 37–44. [DOI] [PubMed] [Google Scholar]
  20. Holli, K. K. , Wäljas, M. , Harrison, L. , Liimatainen, S. , Luukkaala, T. , Ryymin, P. , … Dastidar, P. (2010). Mild traumatic brain injury: Tissue texture analysis correlated to neuropsychological and DTI findings. Academic Radiology, 17(9), 1096–1102. [DOI] [PubMed] [Google Scholar]
  21. Hua, K. , Zhang, J. , Wakana, S. , Jiang, H. , Li, X. , Reich, D. S. , … Mori, S. (2008). Tract probability maps in stereotaxic spaces: Analyses of white matter anatomy and tract‐specific quantification. NeuroImage, 39(1), 336–347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kickingereder, P. , Burth, S. , Wick, A. , Götz, M. , Eidel, O. , Schlemmer, H. , … Radbruch, A. (2016). Radiomic profiling of glioblastoma: Identifying an imaging predictor of patient survival with improved performance over established clinical and radiologic risk models. Radiology, 280(3), 880–889. [DOI] [PubMed] [Google Scholar]
  23. Kraemer, M. , Buerger, M. , & Berlit, P. (2010). Diagnostic problems and delay of diagnosis in amyotrophic lateral sclerosis. Clinical Neurology and Neurosurgery, 112(2), 103–105. [DOI] [PubMed] [Google Scholar]
  24. Leemans, A. , Jeurissen, B. , Sijbers, J. , & Jones, D. (2009). ExploreDTI: A graphical toolbox for processing, analyzing, and visualizing diffusion MR data In Paper presented at the 17th Annual Meeting of Intl Soc Mag Reson Med (Vol. 209, p. 3537). Honolulu, HI: International Society for Magnetic Resonance in Medicine. [Google Scholar]
  25. Maani, R. , Yang, Y. H. , Emery, D. , & Kalra, S. (2016). Cerebral degeneration in amyotrophic lateral sclerosis revealed by 3‐dimensional texture analysis. Frontiers in Neuroscience, 10, 120 10.3389/fnins.2016.00120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Maani, R. , Yang, Y. H. , & Kalra, S. (2015). Voxel‐based texture analysis of the brain. PLoS One, 10(3), e0117759. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Mazziotta, J. , Toga, A. , Evans, A. , Fox, P. , Lancaster, J. , Zilles, K. , … Mazoyer, B. (2001). A probabilistic atlas and reference system for the human brain: International consortium for brain mapping (ICBM). Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 356(1412), 1293–1322. 10.1098/rstb.2001.0915 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Meadowcroft, M. D. , Mutic, N. J. , Bigler, D. C. , Wang, J. , Simmons, Z. , Connor, J. R. , & Yang, Q. X. (2015). Histological–MRI correlation in the primary motor cortex of patients with amyotrophic lateral sclerosis. Journal of Magnetic Resonance Imaging, 41(3), 665–675. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Perrone, D. , Aelterman, J. , Pižurica, A. , Jeurissen, B. , Philips, W. , & Leemans, A. (2015). The effect of Gibbs ringing artifacts on measures derived from diffusion MRI. NeuroImage, 120, 441–455. [DOI] [PubMed] [Google Scholar]
  30. Pierpaoli, C. , Barnett, A. , Pajevic, S. , Chen, R. , Penix, L. , Virta, A. , & Basser, P. (2001). Water diffusion changes in Wallerian degeneration and their dependence on white matter architecture. NeuroImage, 13(6), 1174–1185. [DOI] [PubMed] [Google Scholar]
  31. Pupillo, E. , Messina, P. , Logroscino, G. , & Beghi, E. (2014). Long‐term survival in amyotrophic lateral sclerosis: A population‐based study. Annals of Neurology, 75(2), 287–297. [DOI] [PubMed] [Google Scholar]
  32. Sarica, A. , Cerasa, A. , Valentino, P. , Yeatman, J. , Trotta, M. , Barone, S. , … Pucci, F. (2017). The corticospinal tract profile in amyotrophic lateral sclerosis. Human Brain Mapping, 38(2), 727–739. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Schuster, C. , Hardiman, O. , & Bede, P. (2016). Development of an automated MRI‐based diagnostic protocol for amyotrophic lateral sclerosis using disease‐specific pathognomonic features: A quantitative disease‐state classification study. PLoS One, 11(12), e0167331. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Song, S. , Yoshino, J. , Le, T. Q. , Lin, S. , Sun, S. , Cross, A. H. , & Armstrong, R. C. (2005). Demyelination increases radial diffusivity in corpus callosum of mouse brain. NeuroImage, 26(1), 132–140. [DOI] [PubMed] [Google Scholar]
  35. Sørensen, L. , Igel, C. , Pai, A. , Balas, I. , Anker, C. , Lillholm, M. , … Alzheimer's Disease Neuroimaging Initiative . (2017). Differential diagnosis of mild cognitive impairment and alzheimer's disease using structural MRI cortical thickness, hippocampal shape, hippocampal texture, and volumetry. NeuroImage: Clinical, 13, 470–482. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Verstraete, E. , Veldink, J. H. , Hendrikse, J. , Schelhaas, H. J. , van den Heuvel, M. P. , & van den Berg, L. H. (2012). Structural MRI reveals cortical thinning in amyotrophic lateral sclerosis. Journal of Neurology, Neurosurgery, and Psychiatry, 83(4), 383–388. 10.1136/jnnp-2011-300909 [DOI] [PubMed] [Google Scholar]
  37. Vos, S. B. , Tax, C. M. , Luijten, P. R. , Ourselin, S. , Leemans, A. , & Froeling, M. (2017). The importance of correcting for signal drift in diffusion MRI. Magnetic Resonance in Medicine, 77(1), 285–299. [DOI] [PubMed] [Google Scholar]
  38. Zhang, Y. , Moore, G. , Laule, C. , Bjarnason, T. A. , Kozlowski, P. , Traboulsee, A. , & Li, D. K. (2013). Pathological correlates of magnetic resonance imaging texture heterogeneity in multiple sclerosis. Annals of Neurology, 74(1), 91–99. [DOI] [PubMed] [Google Scholar]
  39. Zhang, Y. , Zhu, H. , Mitchell, J. R. , Costello, F. , & Metz, L. M. (2009). T2 MRI texture analysis is a sensitive measure of tissue injury and recovery resulting from acute inflammatory lesions in multiple sclerosis. NeuroImage, 47(1), 107–111. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure E1 Flow chart depicting a summary of the analytical steps involved in texture analysis. Steps: 1) 22 textures were calculated, 2) redundant textures were removed, 3) white matter‐sensitive textures were identified by correlating the unique textures to DTI metrics, 4) white matter‐sensitive textures were used for voxel‐wise analysis in both the primary and the independent cohorts, 5) texture values from the primary cohort were used to generate a binary regression model, and 6) used to evaluate the diagnostic performance of the white matter‐sensitive textures in both cohorts.

Figure E2: Gray (top) and white (bottom) matter VBM analyses for participants in the primary cohort. Significant differences at p < .001 with a cluster size threshold of 20 or more voxels were found. The results did not survive multiple comparison correction with FDR. VBM results are overlaid on a sample T1W image in neurological orientation. Detailed results are provided in Table E2. FDR = false discovery rate

Table E1: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure 2.

Table E2: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure E1.

Table E3: Table shows the cluster size, T values, and the MNI coordinates of the all significant regions reported in Figure 6.

Click here for additional data file. (287.8KB, docx)

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES