Using histogram analysis of the intrinsic brain activity mapping to identify essential tremor

doi:10.21203/rs.3.rs-1975329/v1

Download PDF

Research Article

Using histogram analysis of the intrinsic brain activity mapping to identify essential tremor

https://doi.org/10.21203/rs.3.rs-1975329/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Essential tremor (ET) is one of the most common movement disorders. Histogram analysis based on brain intrinsic activity imaging has a promising way to identify ET patients from healthy controls (HCs) and further explore the spontaneous brain activity changes and build the potential diagnostic biomarker in ET patients.

Methods

The histogram features extracted from 133 ET patients and 135 well-matched HCs as the input features. Then, a two-sample t-test, the mutual information, and the least absolute shrinkage and selection operator (LASSO) methods were applied to reduce the feature dimensionality. Support vector machine (SVM), logistic regression (LR), random forest (RF), and k-nearest neighbor (KNN) were used to differentiate ET and HCs and the classification performance was compared in the testing set. Moreover, correlation analysis was carried out between the selected histogram features and clinical tremor characteristics.

Results

Each classifier achieved a good classification performance in training and testing sets. The mean accuracy and area under the curve (AUC) of SVM, LR, RF, and KNN in the testing set were 92.62%, 0.948; 92.01%, 0.942; 93.88%, 0.941; 92.27%, 0.939, respectively. The most power discriminative features were mainly located in the cerebello-thalamo-motor and non-motor cortical pathways. Correlation analysis showed that there were 2 histogram features negatively and 1 positively correlated with tremor severity.

Conclusions

Our findings demonstrated that the histogram analysis of the amplitude of low-frequency fluctuation (ALFF) images with multiple machine learning algorithms could identify ET patients from HCs, and help to understand the spontaneous brain activity pathogenesis mechanisms in ET patients.

Essential tremor

Machine learning

Radiomics

Resting-state fMRI

Cerebellum

Essential tremor (ET) is one of the most common neurological disorders characterized by progressive tremor over time [1]. The 2018 consensus statement of the Movement Disorder Society redefined ET as isolated bilateral upper limb action or postural tremor with a duration of at least 3 years [2]. Meantime, ET accompanied by other neurological symptoms such as parkinsonism, ataxia, rest tremor or non-motor symptoms has been relabeled as “ET plus” [2]. This revised definition gave some advantages to including highly homogeneous ET patients for clinical trials. However, it was only based on clinical characteristics, and the etiology, pathology, biology, and pathogenesis of ET, especially the spontaneous brain activity changes are still very unclear.

The amplitude of low-frequency fluctuations (ALFF) analysis of Resting-state functional magnetic resonance imaging (Rs-fMRI) provides a most promising way to explore spontaneous brain activity changes in lots of neuropsychiatric and neurological disorders, including ET [3–7]. Recently, using ALFF analysis very few studies gained some variable and inconsistent findings, and most of these studies supported that changes of ALFF in cerebello-thalamo-cortical circuits were associated with ET patients [4, 8, 9]. However, these studies were traditional mass univariate analyses, and they could not be used to predict ET patients at an individual level. Meantime, these ALFF analysis methods only used the average value of the ALFF image, and the ALFF image actually contained vast numbers of quantitative information, such as the histogram analysis features. Fortunately, these shortcomings can be remedied by radiomics analysis. It can abstract vast quantitative features including histogram information from ALFF images, and then these features are inputted for machine learning (ML) algorithms [10–13]. ML builds the optimal models by learning and training from massive input data and then applies the model to new data to predict and analyze diseases based on a single-subject level [14–16]. Using ML algorithms combined with voxel-level local connectivity or frequency-dependent intrinsic brain activity analysis, our more recent studies revealed that these ML algorithms could achieve good classification performance to identify ET from healthy controls (HCs) [17–19]. However, up to now, no studies has been combined the histogram analysis based on ALFF images with ML algorithms to identify ET patients.

In the present study, we explored whether combined histogram analysis of ALFF images with multiple ML algorithms could be used to identify ET patients from HCs. We expected these classification models could achieve good classification performance, and these brain regions of the most power discriminative features would not only help to reveal the intrinsic brain activity changes but also further acted as the potential diagnosis biomarker in ET patients.

Participants

This study was approved by the Ethics Committee of the First Affiliated Hospital of Chongqing Medical University in accordance with the Declaration of Helsinki. All participants were recruited from the First Affiliated Hospital of Chongqing Medical University and were required to sign the informed consent. Inclusion criteria of participants were the following: 1) All ET patients were diagnosed by two experienced neurologists (OM C, and Z X) according to the 2018 Movement Disorders Consensus Criteria; 2) The patients had an onset age between 18 to 55 years old, and patients with earlier or later onset were not included; 3) All participants were without thyroid disease, Parkinson’s disease, dystonia, psychogenic tremor, stroke, epilepsy, head injury or any other neurological problems, and none of the HCs reported having relatives with ET; 4) All participants met the image quality and head motion control criteria (see Supplementary text 1). Finally, a total of 268 right-handed participants were recruited, including 133 ET patients and 135 age- and sex-matched HCs. In addition, demographic and clinical information was acquired immediately after completion of the MRI examination. The Fahn-Tolosa-Marin Tremor Rating Scale (TRS) and the Essential Tremor Rating Assessment Scale (TETRAS) were used to assess tremor severity and quality of life in ET patients. We also recorded the tremor frequency index from electromyography examination in ET patients. The Mini-Mental State Examination (MMSE), 17-item Hamilton Depression Rating Scale (HDRS-17), and Hamilton Anxiety Rating Scale (HARS-14) were used to briefly assess the cognitive function and mood status of all the participants, and remove of dementia (MMSE < 24), depression (HDRS-17 > 7) and anxiety (HARS-14 > 7).

MRI Data Acquisition

All MRI images were acquired using a GE Signa Hdxt 3.0-T MRI scanner (General Electric Medical Systems, Waukesha, WI) equipped with a standard 8-channel head coil. Rs-fMRI data were acquired using an echo-planar imaging (EPI) sequence with the following scan parameters: repetition time (TR) = 2000ms, echo time (TE) = 40ms, flip angle (FA) = 90°, 33 axial slices, slice thickness/gap = 4.0/0 mm, field of view (FOV) = 240 × 240 mm, matrix = 64 × 64, and a total of 240 volumes were collected which lasted 8 minutes. High-resolution 3D T1-weighted images were acquired using the following parameters: TR = 8.3ms, TE = 3.3ms, FA = 15°, FOV = 240 × 240 mm, matrix = 256 × 192, slice thickness/gap = 1.0/0 mm. And T2-weighted FLAIR images (TR = 8000 ms, TE = 126 ms, TI = 1500 ms, slice thickness/gap = 5.0/1.5mm, FOV = 240 × 240 mm, and matrix = 256 × 192) were also acquired. During the scanning process, participants were told to keep their eyes closed, relax without actively thinking and stay awake in particular. Earplugs and foam padding were used to reduce scanner noise and minimize head movement.

Data preprocessing and ALFF Calculation

Data preprocessing was conducted using the Data Processing and Analysis of Brain Imaging toolbox (DPABI) [20], and the briefly preprocessing steps were as follows: 1) Removal of the first 10 time points. For scanner stabilization and the acclimation of subjects to the MR scanning environment, the first 10 volumes were discarded, and the remaining 230 time points were included in the subsequent data preprocessing; 2) Slice timing correction. This was used to correct for a different acquisition time across slices in a volume; 3) Realignment. This was used to realign the subsequent functional images to the first volume to correct for within-run head motions, resulting in Friston 24 head motion parameters. These parameters were employed to assess the head movement and ensure the quality of Rs-fMRI data; 4) T1 segmentation and spatial normalization. The T1 images were co-registered to the mean rs-fMRI data for each subject. Specifically, 3D T1-weighted images were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) probability maps using SPM DARTEL segmentation. All the GM, WM, and CSF images were resampled to isotropic 1.5-mm voxels, spatially normalized to the MNI space using both affine transformation and non-linear deformation, and later, resampled to isotropic 3-mm voxel resolution with Rs-fMRI, and the deformation field was applied to the Rs-fMRI data; 5) Regression out Friston’s 24 head motion parameters [21] and the mean time series of global, WM and CSF signals; 6) Spatial smoothing with a Gaussian kernel of 4 mm full width at half maximum; and 7) Detrending and filtering. These steps removed the extremely low-frequency drift and the high-frequency physiological noises. For detrending, we used first-order polynomial functions; and for filtering, we adopted band-pass filtering (0.01 Hz ~ 0.08 Hz) to the time series for each voxel. ALFF analysis was based on the pre-processed images. DPABI was also used to compute the ALFF as Zang et al [22] described. Briefly, the filtered time series of each voxel was transformed into the frequency domain by Fast Fourier Transform. Then, the power spectrums of the signal between 0.01 ~ 0.08HZ were calculated, and the square root of the power spectrums were ALFF values. Finally, for reducing the influence of individual ALFF variation and standardization purposes, the ALFF of each voxel was further divided by the global mean of ALFF values for each participant, and then the created individual smALFF maps of each subject for further analysis in our study.

Abstraction of the histogram Features

Previous studies have demonstrated that some nucleus such as the ventral posterior lateral nucleus of the thalamus and cerebellar dentate nucleus were associated with tremor in ET patients [23–25]. For this reason, an automated anatomical labeling atlas 3 (AAL3) with thalamus parcellation and combined bilateral cerebellar dentate nucleus was used to define the regions of interest (ROIs) in our study. However, some structures in the AAL3 altas (resolution: 1 × 1 × 1 mm) are so small (such as the reuniens nucleus of thalamus and the right ventral tegmental area) that they could not be identified in the Rs-fMRI images (resolution: 3 × 3 × 3 mm), and finally, only 159 ROIs were defined for extracting 15 intensity-based histogram features, including the mean, median, maximum, range, variance, skewness, kurtosis, 10th percentile, 90th percentile, inter-quartile range, mean absolute deviation, robust mean absolute deviation, root mean squared, energy and total energy. We extracted a total of 2385 features from each participant’s mALFF images. The feature extraction procedure was performed using the open-source python package pyradiomics, and the description and formula of each feature can be found on their website (https://pyradiomics.readthedocs.io/en/latest/features.html).

Feature Selection

Due to the curse-of-dimensionality or small-n-large-p problem [26], a total of 2385 features greatly exceeded the sample size while most features are redundant and irrelevant. Therefore, feature selection is a necessary step to obtain the most important features and improve the accuracy of the model. Before the feature selection, the dataset was partitioned into training and testing sets in the ratio of 7:3, and a Z-score standardization was performed respectively to keep the data in sets mutually independent. Then, feature selection was conducted in the training set in three steps. Firstly, we performed a two-sample t-test on the 2385 histogram features, and features with P < 0.01 were selected for the subsequent analysis. Then, the mutual information method (threshold = 0.1) was applied to further reduce dimensionality. Finally, we further applied a least absolute shrinkage and selection operator (LASSO) regression model to choose the most important features for classification [27]. The LASSO performed both regularization and variable selection that compresses high-dimensional data by shrinking coefficients for weaker predictors toward zero and dropping variables from the model when their coefficients reach zero. A penalty term (|βi|) is added to the linear regression model in LASSO which can shrink coefficients towards zero (L1 regularization). As the penalty term increases, the Lasso sets more coefficients to zero. The loss function of LASSO is as follows:

The penalization parameter λ was tuned under the criteria of minimal mean squared error (MSE) to construct the optimal subset of features via a 10-fold cross-validated grid-search approach. Features with non-zero coefficients in the LASSO regression model were selected to train the classification model.

Model Construction and Model Evaluation

In our study, we constructed a nested loop to build the model, of the outer loop was applied to evaluate the model performance by splitting the training and testing sets 100 rounds with a 7:3 ratio randomly, while the inner loop was used to determine the best parameters of classifiers via a grid search method with 10-fold cross-validation. The whole procedure of the nested loop was illustrated in Supplementary Fig. 1. We used some common classifiers, the support vector machine with radial basis function kernel (RBF-SVM), logistic regression with the linear kernel (linear-LR, peanlity = L2), random forest (RF), and k-nearest neighbor (KNN) to build models to differentiate ET and HCs based on the selected features. In the training set, a grid search method with 10-fold cross-validation was used to obtain the optimal parameters of each classifier (inner loop). After finding each optimal model, it was fitted to the entire training set and evaluated in the testing set, this process was repeated 100 times (outer loop).

We took the mean value of accuracy (mACC), balance accuracy (mBACC), sensitivity (mSN), specificity (mSP) into account to assess the classification results, while the mean receiver operating characteristic (mROC) curve was plotted to determine the performance of the classifiers, and the mean area under the curve (mAUC) was applied to evaluate the classification performance and validity of the diagnosis of these models. The model with the highest mAUC in the testing set will be considered the best model. In addition, permutation tests (1000 times) were performed to test the significance of model performances by evaluating whether the accuracy and AUC were significantly higher than values by chance [14, 28].

Identification of discriminative features

Due to the training set being slightly different in each outer iteration, the selected features were different, too. Therefore, we considered the features which were selected greater than or equal to 80 times in all 100 iterations as the most power discriminative features and calculated the mean weights according to all outer iterations.

Statistical Analysis

We used SPSS to statistical process and analyze the demographic and clinical characteristics, the Kolmogorov-Smirnov test was used to assess the normality of continuous variables, the two-sample t-test was performed to analyze the normally distributed variables, the Mann-Whitney U test was used to analyze the non-normally distributed variables, and the chi-square test was applied to test the sex. Finally, a partial Pearson’s correlation analysis was conducted to assess the relationships between the most power discriminative features and clinical tremor characteristics in ET patients with the Bonferroni multiple corrections. A P-value < 0.05 was considered statistically significant.

Demographic and Clinical Characteristics

In this study, 133 ET patients and 135 HCs were included. The demographic and clinical characteristics of all participants are summarized in Table 1. The demographic and clinical characteristics between the ET and HCs did not differ, except for the HARS-14 score and MMSE score were significantly different between the two groups (P = 0.0005 and P = 0.0034, respectively).

Table 1

Demographic and clinical features of ET and HCs
Measure	ET	HCs	Statistics	P - value
Demographic
Sample size	133	135	NA	NA
Age (years)	46.43 ± 14.14	44.58 ± 12.86	T = 1.12	0.2633
Gender (M:F)	67:66	78:57	Z= -1.21	0.2249
Education (years)	12.94 ± 4.47	12.25 ± 4.76	T = 1.22	0.2235
Handedness (R/L)	133:0	135:0	Z = 0.00	1.0000
Cigarette smoker	35/133	33/135	Z= -0.35	0.7253
Clinical of tremor
Tremor of onset (years)	33.80 ± 10.60	NA	NA	NA
Tremor of duration (years)	12.62 ± 9.28	NA	NA	NA
Positive family history		NA	NA	NA
Positive	39	NA	NA	NA
Negative	94	NA	NA	NA
Alcohol sensitivity		NA	NA	NA
Positive	57	NA	NA	NA
Negative	41	NA	NA	NA
NA	35	NA	NA	NA
Tremor medication		NA	NA	NA
Propranolol	31(38.71 ± 18.48mg)	NA	NA	NA
Tremor symmetry		NA	NA	NA
R = L	96	NA	NA	NA
R < L	11	NA	NA	NA
R > L	26	NA	NA	NA
Tremor frequency	6.96 ± 2.30	NA	NA	NA
TRS-parts A&B	23.31 ± 7.80	NA	NA	NA
TRS-part C	12.85 ± 6.83	NA	NA	NA
TETRAS	21.44 ± 7.14	NA	NA	NA
TET-ADSL	13.52 ± 7.19	NA	NA	NA
Clinical of psychology and cognitive
HDRS-17	2.13 ± 1.17	2.20 ± 1.25	T = -0.44	0.6629
HARS-14	2.95 ± 1.19	2.28 ± 1.81	T = 3.51	0.0005
MMSE	28.66 ± 1.28	29.12 ± 1.25	T = -2.96	0.0034
Head movement
FD_power	0.01 ± 0.06	0.01 ± 0.06	T = -0.09	0.9323
Scrubbing volumes	15.27 ± 7.95	15.77 ± 9.31	T = -0.47	0.6379
ET: essential tremor, HCs: healthy controls, HDRS-17: 17-item Hamilton Depression Rating Scale, MMSE: Mini-Mental State Examination, HARS-14: 14-item Hamilton Anxiety Rating Scale, TRS: Fahn-Tolosa-Marin Tremor Rating Scale.

Classification Performance

Table 2 and Fig. 1 show that the four classifiers including RBF-SVM, linear-LR, RF, and KNN achieved good classification performance in both the training and testing sets, implying the histogram radiomics analysis can effectively distinguish ET patients and HCs. In the training set, the mean accuracy and mAUC of the RBF-SVM, linear-LR, RF, and KNN are 94.87%, 0.970; 93.91%, 0.961; 99.49%, 0.995; 95.16% 0.979, respectively. Despite the superior performance of all methods, we used in the training set, what really matters is the predictive results in the testing set. In the testing set, the mean accuracy and mAUC of the RBF-SVM, linear-LR, RF, and KNN are 92.62%, 0.948; 92.01%, 0.942; 93.88%, 0.941; 92.27%, 0.939, respectively. The RBF-SVM classifier with the highest mAUC of 0.948 in the testing set was considered the best classifier, with the mean accuracy, and balance accuracy was 92.62%, and 92.56%, the mean sensitivity was 87.95%, and mean specificity was 97.17%. Permutation tests were performed in each testing set, and the results indicated the reliability of the observed accuracy and AUC, with all P-values < 0.001 in iteration.

Table 2

The classification performance in the training set and validation set.
Method			Training set				Testing set
	mAUC	mSN(%)	mSP(%)	mACC(%)	mbACC(%)	mAUC	mSN(%)	mSP(%)	mACC(%)	mbACC(%)
SVM	0.970	89.91 ± 2.57	99.78 ± 4.68	94.87 ± 1.33	94.85 ± 1.34	0.948	87.95 ± 4.39	97.17 ± 3.71	92.62 ± 2.74	92.56 ± 2.74
LR	0.961	88.47 ± 2.07	99.30 ± 0.80	93.91 ± 1.22	93.89 ± 1.22	0.942	87.17 ± 4.50	96.73 ± 3.44	92.01 ± 2.54	91.95 ± 2.55
RF	0.995	98.99 ± 1.31	99.99 ± 1.06	99.49 ± 0.66	99.49 ± 0.66	0.941	89.25 ± 4.35	98.39 ± 1.98	93.88 ± 2.44	93.82 ± 2.46
KNN	0.979	90.59 ± 4.66	99.68 ± 0.57	95.16 ± 2.41	95.14 ± 2.42	0.939	87.18 ± 4.48	97.24 ± 2.73	92.27 ± 2.39	92.21 ± 2.40
The values are expressed as mean ± standard deviation (x̄ ± s); m:mean; AUC: area under the curve; SN: sensitivity; SP: specificity; ACC: accuracy; bACC: balance accuracy; SVM: support vector machine; LR: logistic regression; RF: random forest; KNN: k-nearest neighbor

Discriminative Features

The most power discriminative features were identified as features that were repeatedly selected ≥ 80% in the outer iteration, and the brain regions where these features are located are considered to be helpful for ET classification. A total of 19 features are reported (Table 3, Figs. 2 and 3), including the left cerebellar lobule III (Cerebellum_3_L)_kurtosis, left cerebellar lobule IV ~ V with two features, (Cerebellum_4_5_L)_mean, (Cerebellum_4_5_L)_kurtosis, respectively; right cerebellar lobule IV ~ V with four features, (Cerebellum_4_5_R)_mean, (Cerebellum_4_5_R)_{total energy}, (Cerebellum_4_5_R)_kurtosis, (Cerebellum_4_5_R)_{90th percentile}, respectively; right cerebellar lobule Ⅷ (Cerebellum_8_R)_{90 Percentile}; left medial mediodorsal nucleus of thalamus (Thal_MDm_L)_kurtosis; left lateral mediodorsal nucleus of thalamus (Thal_MDl_L)_variance; left ventral posterior lateral nucleus of thalamus (Thal_VPL_L)_kurtosis; right ventral posterior lateral nucleus of thalamus (Thal_VPL_R)_kurtosis; right precentral gyrus (Precentral_R)_kurtosis; left medial superior frontal gyrus (Frontal_Sup_Med_L)_kurtosis; right medial superior frontal gyrus (Frontal_Sup_Med_R)_kurtosis; left insula (Insula_L)_kurtosis; right insula (Insula_R)_kurtosis; left supplementary motor area (Supp_Motor_Area_L)_kurtosis; left dentate nucleus (dentate_L)_mean.

Table 3

The significant discriminative features between ET and HCs.
AAL3 number	Features	AAL3 brain areas	Coefficient	Frequency
102	Total Energy	Right cerebellar lobule IV ~ V	0.190466842	95
33	Kurtosis	Left Insula	0.034522173	98
129	Kurtosis	Left ventral posterior lateral nucleus of thalamus	0.032302957	90
2	Kurtosis	Right precentral gyrus	0.030070535	99
20	Kurtosis	Right medial superior frontal gyrus	0.029236559	99
34	Kurtosis	Right Insula	0.025480698	99
15	Kurtosis	Left supplementary motor area	0.021742021	90
137	Variance	Left lateral mediodorsal nucleus of thalamus	0.021271266	82
135	Kurtosis	Left medial mediodorsal nucleus of thalamus	0.021267412	99
19	Kurtosis	Left medial superior frontal gyrus	0.019697482	98
130	Kurtosis	Right ventral posterior lateral nucleus of thalamus	0.014803541	80
99	Kurtosis	Left cerebellar lobule III	-0.020707327	87
101	Kurtosis	Left cerebellar lobule IV ~ V	-0.021590975	82
102	Kurtosis	Right cerebellar lobule IV ~ V	-0.023882283	85
	Mean	Left dentate nucleus	-0.044511876	89
108	90th Percentile	Right cerebellar lobule VIII	-0.062860398	84
102	90th Percentile	Right cerebellar lobule IV ~ V	-0.078035109	81
101	Mean	Left cerebellar lobule IV ~ V	-0.078885995	98
102	Mean	Right cerebellar lobule IV ~ V	-0.275559008	100
AAL3: Automated Anatomical Labeling 3

Correlations between the discriminative features and clinical tremor symptoms

Figure 4 showed the partial Pearson’s correlation analysis results, and 3 histogram features were significantly correlated with clinical tremor characteristics in ET patients. The kurtosis of left cerebellar lobule IV ~ V and the total energy of right cerebellar lobule IV ~ V were negative correlation with TRS part A&B (P < 0.001, r = − 0.42 and − 0.43 respectively), and the kurtosis of right precentral gyrus was positive correlation with TRS part A&B (P < 0.001, r = 0.54).

In the present study, combined the histogram features based on the intrinsic brain activity mapping with multiple machine learning algorithms to identify ET patients from HCs, and three main findings were reported. First, all of the four machine learning algorithms including RBF-SVM, linear-LR, RF, and KNN achieved good classification performances; second, the most power discriminative histogram features were mainly located in the cerebello-thalamo-motor and non-motor cortical pathways; third, some histogram features could be used to explain partially clinical tremor symptoms.

Histogram analysis as one of the most common radiomics methods is a newly emerging area in quantitative image analysis, and in this process, the medical images were divided into amounts of quantitative features. On account of this advantage, a few studies have achieved good classification performance for identifying Parkinson’s disease from HCs or Alzheimer's Disease from HCs [12, 28–30]. Meanwhile, due to the shines of predicting the individual subject and the multivariate nature of machine learning algorithms, radiomics analysis has been successfully used for neurological disease research and provided quantitative and objective support for clinical diagnosis and disease prognosis [29, 31, 32]. Combined clinical symptoms or brain gray matter volumes with machine learning algorithms, few studies achieved good classification performance in classifying ET from HCs [33–36]. Consistent with the above studies, our studies also achieved good classification performance. Meantime, our results revealed that among the top 19 most power discrimination features, only three discrimination features were mean ALFF values, and the other histogram analysis information were the total energy, kurtosis, variance, and 90th percentile. These aspects further suggested that compare with the traditional ALFF analysis, the histogram analysis could give more quantitative information, and it may be more susceptible to revealing ALFF changes in ET patients.

In this study, the most power discriminative features were mainly located in the cerebello-thalamo-motor cortical pathways, including the cerebellum, thalamus, precentral gyrus, and the supplementary motor area, and it was consistent with the classical tremor network theories. Using traditional ALFF analysis, very few studies also revealed changes ALFF in the classical tremor network were related to ET patients. However, these results were variable and even contradictory, and all of these studies did not reveal ALFF changes in the thalamus. Yin, et al [37] revealed that decreased ALFF in the cerebellum and increased ALFF in cerebral cortices were associated with ET patients. Li, et al [38] gain a contradictory result, increased ALFF in the cerebellum and decreased ALFF in cerebral cortices were involved in ET patients. Our results seemed to be contradictory to previous studies, and we speculated that the following reasons may be reasonable explanations. First, ET as etiological, clinical, and pathological heterogeneity diseases, the heterogeneous properties may cause the variable results from different researchers. Second, due to the small sample and absence of strict inclusion criteria, these results were more variable. Third, the ventral posterior lateral nucleus of the thalamus could not be identified in the common atlas such as the anatomical automatic labeling atlas (AAL), Harvard Oxford atlas (HOA), and Brainnetome atlas (BNA) et al, and these caused difficulty in directly revealing the ALFF changes in the classical tremor network. Finally, all the above studies also revealed ALFF changes in the cerebellar-cortical network. Therefore, our results were actually in line with the previous studies. Meantime, compare with the previous studies, a large sample size (133 ET patients and 135 HCs) and a strict inclusion (the 2018 Consensus Criteria of the Movement Disorder Society) were adopted in our studies. So, we suggested that the most power discriminative features located in the classical tremor network further reinforced the classical tremor network pathogenesis theories, and decreased histogram analysis matrices in cerebellum such as total energy, kurtosis, mean and 90th percentile maybe reflected the primary pathological injury of the cerebellum in ET patients.

Moreover, the most power discriminative features were not only confined to the classical tremor network but also extended to the cerebello-non-motor cortical circuits, including the bilateral superior frontal gyrus and insula, and it seemed difficult for us to understand these aspects. First, growing evidence pointed out ET is a syndrome caused by primary pathological damage in the cerebellum. The cerebellums have extensive connectivity with cerebral cortices including motor and non-motor cortices and regulate motor and non-motor functions. Second, strict inclusion criteria without gross cognitive impairment, depression, and anxiety were adopted to gain a highly homogeneous ET cohort in our study. These strict inclusion criteria could not remove of development the above non-motor symptoms in the future, and even get rid of a compensatory state to prevent the development of these non-motor symptoms.

Limitations

There are several limitations of this study. First, although the relatively larger sample size and a good classification performance were achieved in our study, a multi-center data would let our results more generalization and stable in the future. Second, only histogram analysis of intrinsic brain activity mapping was selected as the input feature, and combined multimodal imaging data with clinical metrics perhaps gave a more precise performance in the future. Finally, the diagnosis of ET relied only on clinical symptoms and neurological examinations. Due to the absence of diagnostic biomarkers and the misdiagnosis are common, we adopted strict enrollment criteria and annual follow-up to reduce misdiagnoses.

In this study, combining histogram analysis of ALFF images with multiple machine learning algorithms achieved good classification performance for identifying ET patients from HCs. The most power discriminatory features were not only confined to the typical tremor networks but also extended into non-motor networks, and these features would help to understand the spontaneous brain activity pathogenesis mechanisms in ET patients.

Author Contribution Pan Xiao and Li Tao wrote the main manuscript text and prepared formal analysis, investigation, data curation, data analysis, methodology, software, and visualization. Xiaoyu Zhang, Qin Li, Hong’e Gui, Bintao Xu, and Xueyan Zhang performed the formal analysis, investigation, data curation, and data analysis. Wanlin He, Huiyue Chen, Hansheng Wang, Fajin Lv, Tianyou Luo, Jing Luo, Yun Man, and Zheng Xiao provided the data supporting and critically revised the work. Weidong Fang performed funding acquisition, project administration, supervision, validation, and drafted and critically revised the work. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding This study was funded by the National Natural Science Foundation of China (NSFC: 81671663) and the Natural Science Foundation of Chongqing (NSFCQ: cstc2014jcyjA10047).

Ethics approval Approval was obtained from the ethics committee of the First Affiliated Hospital of Chongqing Medical University. The procedures used in this study adhere to the tenets of the Declaration of Helsinki.

Conflict of Interest The authors declare no competing interests.

Data availability statement The raw data that support the findings of this study are available from the corresponding author, upon reasonable request.

Consent to participate Informed consent was obtained from all individual participants included in the study.

J.K. Wong, C.W. Hess, L. Almeida, E.H. Middlebrooks, E.A. Christou, E.E. Patrick, A.W. Shukla, K.D. Foote, M.S. Okun, Deep brain stimulation in essential tremor: targets, technology, and a comprehensive review of clinical outcomes, Expert Rev Neurother 20(4) (2020) 319–331.
K.P. Bhatia, P. Bain, N. Bajaj, R.J. Elble, M. Hallett, E.D. Louis, J. Raethjen, M. Stamelou, C.M. Testa, G. Deuschl, Consensus Statement on the classification of tremors. from the task force on tremor of the International Parkinson and Movement Disorder Society, Movement Disorders: Official Journal of the Movement Disorder Society 33(1) (2018) 75–87.
Y.-F. Zang, Y. He, C.-Z. Zhu, Q.-J. Cao, M.-Q. Sui, M. Liang, L.-X. Tian, T.-Z. Jiang, Y.-F. Wang, Altered baseline brain activity in children with ADHD revealed by resting-state functional MRI, Brain Dev 29(2) (2007) 83–91.
J.-Y. Li, X.-L. Suo, N.-N. Li, D. Lei, Z.-J. Lu, L. Wang, J.-X. Peng, L.-R. Duan, X. Jing, J. Yi, Q.-Y. Gong, R. Peng, Altered spontaneous brain activity in essential tremor with and without resting tremor: a resting-state fMRI study, MAGMA 34(2) (2021) 201–212.
X. Liu, W. Qian, R. Jin, X. Li, K.D. Luk, E.X. Wu, Y. Hu, Amplitude of Low Frequency Fluctuation (ALFF) in the Cervical Spinal Cord with Stenosis: A Resting State fMRI Study, PLoS One 11(12) (2016) e0167279.
Y. Zhi, Y. Yuan, Q. Si, M. Wang, Y. Shen, L. Wang, H. Zhang, K. Zhang, The Association between DRD3 Ser9Gly Polymorphism and Depression Severity in Parkinson's Disease, Parkinsons Dis 2019 (2019) 1642087.
N. Zhao, L.-X. Yuan, X.-Z. Jia, X.-F. Zhou, X.-P. Deng, H.-J. He, J. Zhong, J. Wang, Y.-F. Zang, Intra- and Inter-Scanner Reliability of Voxel-Wise Whole-Brain Analytic Metrics for Resting State fMRI, Front Neuroinform 12 (2018) 54.
L. Wang, D. Lei, X. Suo, N. Li, Z. Lu, J. Li, J. Peng, Q. Gong, R. Peng, Resting-state fMRI study on drug-naive patients of essential tremor with and without head tremor, Scientific Reports 8(1) (2018) 10580.
W. Yin, W. Lin, W. Li, S. Qian, X. Mou, Resting State fMRI Demonstrates a Disturbance of the Cerebello-Cortical Circuit in Essential Tremor, Brain Topogr 29(3) (2016) 412–418.
X. Cao, X. Wang, C. Xue, S. Zhang, Q. Huang, W. Liu, A Radiomics Approach to Predicting Parkinson's Disease by Incorporating Whole-Brain Functional Activity and Gray Matter Structure, Front Neurosci 14 (2020) 751.
Z. Liu, S. Wang, D. Dong, J. Wei, C. Fang, X. Zhou, K. Sun, L. Li, B. Li, M. Wang, J. Tian, The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges, Theranostics 9(5) (2019) 1303–1322.
D. Shi, H. Zhang, S. Wang, G. Wang, K. Ren, Application of Functional Magnetic Resonance Imaging in the Diagnosis of Parkinson's Disease: A Histogram Analysis, Front Aging Neurosci 13 (2021) 624731.
P. Lambin, E. Rios-Velazquez, R. Leijenaar, S. Carvalho, R.G.P.M. van Stiphout, P. Granton, C.M.L. Zegers, R. Gillies, R. Boellard, A. Dekker, H.J.W.L. Aerts, Radiomics: extracting more information from medical images using advanced feature analysis, Eur J Cancer 48(4) (2012) 441–446.
Z.Y. Tian, L. Qian, L. Fang, X.H. Peng, X.H. Zhu, M. Wu, W.Z. Wang, W.H. Zhang, B.Q. Zhu, M. Wan, X. Hu, J. Shao, Frequency-Specific Changes of Resting Brain Activity in Parkinson's Disease: A Machine Learning Approach, Neuroscience 436 (2020) 170–183.
L. Jollans, R. Boyle, E. Artiges, T. Banaschewski, S. Desrivières, A. Grigis, J.-L. Martinot, T. Paus, M.N. Smolka, H. Walter, G. Schumann, H. Garavan, R. Whelan, Quantifying performance of machine learning methods for neuroimaging data, NeuroImage 199 (2019) 351–365.
M. Khosla, K. Jamison, G.H. Ngo, A. Kuceyeski, M.R. Sabuncu, Machine learning in resting-state fMRI analysis, Magn Reson Imaging 64 (2019) 101–121.
X. Zhang, L. Tao, H. Chen, X. Zhang, H. Wang, W. He, Q. Li, F. Lv, T. Luo, J. Luo, Y. Man, Z. Xiao, J. Cao, W. Fang, Combined Intrinsic Local Functional Connectivity With Multivariate Pattern Analysis to Identify Depressed Essential Tremor, Frontiers In Neurology 13 (2022) 847650.
X. Zhang, H. Chen, L. Tao, X. Zhang, H. Wang, W. He, Q. Li, P. Xiao, B. Xu, H. Gui, F. Lv, T. Luo, Y. Man, Z. Xiao, W. Fang, Combined multivariate pattern analysis with frequency-dependent intrinsic brain activity to identify essential tremor, Neuroscience Letters 776 (2022) 136566.
Y. Li, L. Tao, H. Chen, H. Wang, X. Zhang, X. Zhang, X. Duan, Z. Fang, Q. Li, W. He, F. Lv, J. Luo, Z. Xiao, J. Cao, W. Fang, Identifying Depressed Essential Tremor Using Resting-State Voxel-Wise Global Brain Connectivity: A Multivariate Pattern Analysis, Front Hum Neurosci 15 (2021) 736155.
C.-G. Yan, X.-D. Wang, X.-N. Zuo, Y.-F. Zang, DPABI: Data Processing & Analysis for (Resting-State) Brain Imaging, Neuroinformatics 14(3) (2016) 339–351.
C.-G. Yan, R.C. Craddock, Y. He, M.P. Milham, Addressing head motion dependencies for small-world topologies in functional connectomics, Frontiers In Human Neuroscience 7 (2013) 910.
Y.F. Zang, Y. He, C.Z. Zhu, Q.J. Cao, M.Q. Sui, M. Liang, L.X. Tian, T.Z. Jiang, Y.F. Wang, Altered baseline brain activity in children with ADHD revealed by resting-state functional MRI, Brain Dev 29(2) (2007) 83–91.
S. Tikoo, S. Pietracupa, S. Tommasin, M. Bologna, N. Petsas, K. Bharti, A. Berardelli, P. Pantano, Functional disconnection of the dentate nucleus in essential tremor, J Neurol 267(5) (2020) 1358–1367.
A.W. Buijink, M. Broersma, A.M. van der Stouwe, G.A. van Wingen, P.F. Groot, J.D. Speelman, N.M. Maurits, A.F. van Rootselaar, Rhythmic finger tapping reveals cerebellar dysfunction in essential tremor, Parkinsonism Relat Disord 21(4) (2015) 383–8.
S. Pietracupa, M. Bologna, S. Tommasin, A. Berardelli, P. Pantano, The Contribution of Neuroimaging to the Understanding of Essential Tremor Pathophysiology: a Systematic Review, Cerebellum (2021).
B. Zhang, S. Liu, X. Liu, S. Chen, Y. Ke, S. Qi, X. Wei, D. Ming, Discriminating subclinical depression from major depression using multi-scale brain functional features: A radiomics analysis, J Affect Disord 297 (2022) 542–552.
R. Tibshirani, Regression shrinkage and selection via the lasso, Journal of the Royal Statistical Society: Series B (Methodological) 58(1) (1996) 267–288.
D. Shi, X. Yao, Y. Li, H. Zhang, G. Wang, S. Wang, K. Ren, Classification of Parkinson's disease using a region-of-interest- and resting-state functional magnetic resonance imaging-based radiomics approach, Brain Imaging Behav (2022).
H. Pang, Z. Yu, R. Li, H. Yang, G. Fan, MRI-Based Radiomics of Basal Nuclei in Differentiating Idiopathic Parkinson's Disease From Parkinsonian Variants of Multiple System Atrophy: A Susceptibility-Weighted Imaging Study, Front Aging Neurosci 12 (2020) 587250.
Q. Feng, M. Wang, Q. Song, Z. Wu, H. Jiang, P. Pang, Z. Liao, E. Yu, Z. Ding, Correlation Between Hippocampus MRI Radiomic Features and Resting-State Intrahippocampal Functional Connectivity in Alzheimer's Disease, Front Neurosci 13 (2019) 435.
Y. Wang, K. Sun, Z. Liu, G. Chen, Y. Jia, S. Zhong, J. Pan, L. Huang, J. Tian, Classification of Unmedicated Bipolar Disorder Using Whole-Brain Functional Activity and Connectivity: A Radiomics Analysis, Cereb Cortex 30(3) (2020) 1117–1128.
D. Shi, H. Zhang, G. Wang, S. Wang, X. Yao, Y. Li, Q. Guo, S. Zheng, K. Ren, Machine Learning for Detecting Parkinson's Disease by Resting-State Functional Magnetic Resonance Imaging: A Multicenter Radiomics Analysis, Front Aging Neurosci 14 (2022) 806828.
J. Yang, D. Lei, J. Peng, X. Suo, W.H.L. Pinaya, W. Li, J. Li, G.J. Kemp, R. Peng, Q. Gong, Disrupted brain gray matter networks in drug-naive participants with essential tremor, Neuroradiology 63(9) (2021) 1501–1510.
A. Suppa, F. Asci, G. Saggio, P. Di Leo, Z. Zarezadeh, G. Ferrazzano, G. Ruoppolo, A. Berardelli, G. Costantini, Voice Analysis with Machine Learning: One Step Closer to an Objective Diagnosis of Essential Tremor, Mov Disord 36(6) (2021) 1401–1410.
S. Moon, H.-J. Song, V.D. Sharma, K.E. Lyons, R. Pahwa, A.E. Akinwuntan, H. Devos, Classification of Parkinson's disease and essential tremor based on balance and gait characteristics from wearable motion sensors via machine learning techniques: a data-driven approach, J Neuroeng Rehabil 17(1) (2020) 125.
S. Shahtalebi, S.F. Atashzar, R.V. Patel, M.S. Jog, A. Mohammadi, A deep explainable artificial intelligent framework for neurological disorders discrimination, Sci Rep 11(1) (2021) 9630.
W. Yin, W. Lin, W. Li, S. Qian, X. Mou, Resting State fMRI Demonstrates a Disturbance of the Cerebello-Cortical Circuit in Essential Tremor, Brain topography 29(3) (2016) 412–8.
J.Y. Li, X.L. Suo, N.N. Li, D. Lei, Z.J. Lu, L. Wang, J.X. Peng, L.R. Duan, X. Jing, J. Yi, Q.Y. Gong, R. Peng, Altered spontaneous brain activity in essential tremor with and without resting tremor: a resting-state fMRI study, Magma 34(2) (2021) 201–212.

No competing interests reported.

Supplementarytext1.doc

Download PDF

Version 1

posted

You are reading this latest preprint version

Using histogram analysis of the intrinsic brain activity mapping to identify essential tremor

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Materials And Methods

Participants

MRI Data Acquisition

Data preprocessing and ALFF Calculation

Abstraction of the histogram Features

Feature Selection

Model Construction and Model Evaluation

Identification of discriminative features

Statistical Analysis

Results

Demographic and Clinical Characteristics

Classification Performance

Discriminative Features

Correlations between the discriminative features and clinical tremor symptoms

Discussion

Limitations

Conclusions

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1